Geography and Economic Development

John Luke Gallup and Jeffrey D. Sachswith

Andrew D. Mellinger

keywords: geography, empirical growth models, transportation costs, tropical disease, tropical agriculture, urbanization,population

JEL codes:O10 Economic Development - General

O13 Agriculture and Natural ResourcesO40 Economic Growth and Aggregate ProductivityO57 Comparative Economywide Country Studies

This research relies heavily on the work of a group at the Center for International Development, especially Steven Radelet’s workon the role of transport costs, and Andrew Warner’s work on natural resource abundance. Our progress in exploring the role ofgeography and the long process of creating new datasets has been shared with these researchers. We thank Vernon Henderson,Jean-Claude Milleron, Boris Pleskovic , and Anthony Venables for helpful comments. This research was partially funded by theOffice of Emerging Markets, Economic Growth Center, Bureau for Global Programs, Field Support and Research, United StatesAgency for International Development, under the Consulting Assistance for Economic Reform (CAER) II Project, Contract No.PCE-C-00-95-00015-00.

Geography and Economic Development

John Luke Gallup and Jeffrey D. Sachswith

Andrew D. Mellinger

Abstract

This paper addresses the complex relationship between geography and macroeconomic growth. We investigate the ways in which geography may matter directly for growth, controlling foreconomic policies and institutions, as well as the effects of geography on policy choices andinstitutions. We find that location and climate have large effects on income levels and incomegrowth, through their effects on transport costs, disease burdens, and agricultural productivity,among other channels. Furthermore, geography seems to be a factor in the choice of economicpolicy itself. When we identify geographical regions that are not conducive to modern economicgrowth, we find that many of these regions have high population density and rapid populationincrease. This is especially true of populations that are located far from the coast, and thus thatface large transport costs for international trade, as well as populations in tropical regions of highdisease burden. Furthermore, much of the population increase in the next thirty years is likely totake place in these geographically disadvantaged regions.

John Luke Gallup is a Research Fellow at the Center for International Development, an Institute Associate at the HarvardInstitute for International Development, and a Lecturer in the Department of Economics at Harvard University. His recent researchfocuses on economic geography, malaria, and the relationship between economic growth and poverty alleviation.

Jeffrey D. Sachs is the Director of the Center for International Development and the Harvard Institute for InternationalDevelopment, and the Galen L. Stone Professor of International Trade in the Department of Economics. His current researchinterests include emerging markets, global competitiveness, economic growth and development, transition to a market economy,international financial markets, international macroeconomic policy coordination and macroeconomic policies in developing anddeveloped countries.

Andrew D. Mellinger is a Research Associate at the Center for International Development specializing in geographic informationsystems applications. His recent research includes methods of sustainable planning, the impact of urbanization on bio-diversity,and modeling the landscape ecological effects of road networks and logging regimes.

3

I. Introduction

Two centuries after the start of modern economic growth, a large portion of the worldremains mired in poverty. Some benefits of modern development, especially gains in lifeexpectancy and reduced infant mortality, have spread to nearly all parts of the world, though hugeand tragic discrepancies remain in even these areas. In material well being, however, as measuredby gross domestic product per capita adjusted for purchasing power parity (PPP), the yawninggaps are stunning and show few signs of amelioration. According to the valuable data assembledby Angus Maddison for the Organization of Economic Cooperation and Development (1995),Western Europe outpaced Africa in average per capita GDP by a factor of around 2.9 in 1820, andby a factor of 13.2 by 1992. More stunningly, Madison puts the African per capita income in 1992at $1,284 dollars (measured in 1990 PPP adjusted dollars), which is essentially identical toMaddison’s estimate of the average GDP per capita in Western Europe in 1820, $1,292. One areaof the developing world, Asia, showed significant progress during the past thirty years, withaverage incomes rising from around $1,212 in 1965 to $3,239 in 1992 on the Maddison data.1 InAfrica, however, the levels of income in the 1990s were about the same as in 1970. (Maddisonputs Africa’s average income at $1,289 in 1971 and $1,284 in 1992). In Latin America and theCaribbean, average income levels in 1992 ($4,820) were only 6.6 percent higher than in 1974($4,521).

Figure 1 shows the global map of GDP per capita as of 1995 (using PPP-adjustedestimates from World Bank 1997). Two geographical correlates of economic development areunmistakable. First, the countries in the geographical tropics are nearly all poor.2 Almost allhigh-income countries are in the mid- and high latitudes. Second, coastal economies are generallyhigher income than the landlocked economies. Indeed, outside of Europe, there is not a singlehigh-income landlocked country, though there are 29 non-European landlocked countries.

Figure 2 shows the distribution of population density, measured as populationper km2. Unlike GDP, for which data generally are available only on a national or broad sub-national basis, the world distribution of population as of 1994 is now available on a geographicinformation system (GIS) basis, with a resolution of 5 minutes by 5 minutes.3 We can immediatelymake three observations about population density. First, there is no simple relationship betweenpopulation density and income level. We find densely populated regions that are rich (WesternEurope) and poor (India, Indonesia, and China), and sparsely populated regions that are both rich(Australia and New Zealand) and poor (the Sahel of Africa). On a cross-country basis, there is aweak positive correlation between population density and GDP per capita.4 Second, the greatEurasian landmass is more densely populated than the rest of the world. (This seems to be a

1These figures refer to the unweighted average of GDP per capita of nine countries: China, Korea, Taiwan, Hong Kong,Singapore, the Philippines, Malaysia, Indonesia, and Thailand.

2 The geographical tropics refers to the area between the Tropic of Cancer (23.45 N latitude) and the Tropic ofCapricorn (23.45 S latitude), the band in which the sun is directly overhead at some point during the year.

3All data are described in the data appendix. While the population data are presented on a 5 minute by 5 minuterefinement, some of the underlying data base is actually less refined, with population interpolated to the 5 minute by 5 minute grid.

4For the universe of 150 countries with population greater than 1 million, the correlation between population density(population per km2) and GDP per capita in 1995 is 0.32.

4

function of human history in addition to underlying geophysical and biogeographical conditions, aswe show later). Third, the coastlines and areas connected to the coast by navigable rivers aremore densely populated than the hinterlands (we will use this term to refer to regions more than100 km from the coast or an ocean-navigable waterway). Part of our goal is to decipher some ofthe sources of population density, and the somewhat subtle relationship of population density toincome levels.

If we multiply GDP per capita and population density, we can calculate GDP density,measured as GDP per km2, shown in Figure 3. In line with the first two figures, the coastal,temperate, Northern-hemisphere economies have the highest economic densities in the world. Four of these areas — Western Europe, Northeast Asia (coastal China, Japan, and Korea), and theEastern and Western seaboards of the U.S. and Canada — are the core economic zones of themodern world.5 These regions are the overwhelming providers of capital goods in global trade,the world’s financial centers, and the generators of a large proportion of global production. If wetake the regions within the U.S., Western Europe, and temperate-zone East Asia that lie within 100km of the coastline, these areas account for a mere 3 percent of the world’s inhabited land area, 13percent of the world’s population, and at least 32 percent of the world’s GDP measured atpurchasing power parity.6 If we exclude coastal China from the calculations, which lags farbehind the other economies in this group, then the core coastal region has a mere 9 percent of theworld’s population but produces at least 30 percent of the world GDP. According to recent dataof the World Trade Organization (1995), just 11 countries in North America, Western Europe, andEast Asia, with 14 percent of the world’s population, account for remarkable 88 percent of globalexports of capital goods (machinery and transport equipment).7

To take a closer look at these patterns, we examine the per capita GDPs of all 150countries in the world with population of 1 million or more in 1995. In total, these 150 countrieshad a combined 1995 population of 5.65 billion, out of a global population estimated to be 5.67billion. Therefore, our universe of observation includes 99.7 percent of the world population. For purposes of discussion, we define a tropical country as one in which half or more of the landarea is within the geographical tropics. There are 72 tropical countries, with 41 percent of theworld population, and 78 non-tropical countries, with 59 percent of the world population. Amongthe tropical countries, the simple average of 1995 GDP per capita (not weighted by countrypopulation) is $3,326. Among the non-tropical countries, the average is $9,027, or nearly threetimes greater. A simple test of the difference of means across the two groups is significant at the p

5For these purposes we include the U.S. and Canadian regions bordering the ocean-navigable Saint Lawrence Seaway

and Great Lakes.6We have not yet assembled sub-national GDP data. Therefore, to make the calculation we assume that GDP per

capita is identical in all regions within a country. This understates the size of GDP in the coastal regions, since GDP per capitatends to be higher in coastal regions.

7These 11 countries, in declining order of shares in global exports, are as follows (with shares in parentheses): U.S.(20.0), Japan (20.0), Germany (14.6), France (6.4), U.K. (5.8), Italy (4.9), Canada (4.7), Korea (3.2), Taiwan (2.9), Belgium(2.7), Netherlands (2.5). Other major exporting countries that are closely linked to the core production system include: Singapore(4.3), China (2.7), Mexico (2.3), Malaysia (2.2), and Hong Kong (0.7).

5

< .001 level.8 Using GIS based calculations to allocate populations within countries that are partlytropical and partly non-tropical, we calculate that 1,804 million live in the geographical tropics,or approximately 32 percent of the world’s population.

It is convenient to divide the non-tropical countries into two sub-groups, the temperate-zone economies and the sub-tropical economies. For our purposes, we define sub-tropical ascountries in which half or more of the land area is tropical or sub-tropical ecological zones, but inwhich the country’s land area is more than half outside of the geographical tropics.9 There are 15sub-tropical economies, and 63 temperate-zone economies. While the tropical countries havemean income of $3,326, the sub-tropical countries have mean income of $7,874, and thetemperate-zone economies have a mean income of $9,302. Among the economies that were notsocialist in the postwar period, the geographical divide is even sharper: non-socialist tropics,$3,685; non-socialist sub-tropics, $9,248; and non-socialist temperate, $14,828.

Of the top thirty countries ranked by 1995 PPP-adjusted GDP per capita, only two aretropical, and these two are tiny: Hong Kong and Singapore. Four are subtropical, and 23 aretemperate-zone. The two tropical countries account for a mere 1.0 percent of the combinedpopulation of the top 30 countries. Using geographic information system data, we can alsoexamine the proportion of the population living in the geographical tropics in the top 30 countries,taking into account that four of the top thirty countries that we have not counted as tropical(Australia, Chile, Taiwan, and the United Arab Emirates) have a part of their populations in thetropical region. Making this adjustment, the tropical share of the top-30 population is 2.3 percent.

Nearly all landlocked countries in the world are poor, except for a handful in Western andCentral Europe which are deeply integrated into the regional European market, and connected bylow-cost trade.10 (Even mountainous Switzerland has the vast bulk of its population in the low-elevation cantons north of the Alps, and these population centers are easily accessible to the NorthAtlantic by land and river-based traffic.). There are 35 landlocked countries in the world withpopulation greater than 1 million, of which 29 are outside of Western and Central Europe. Ofthese 29 countries, the richest is 38th (!), Botswana, which owes it pride of place to well-manageddiamond mines. The second richest is 68th, Belarus. The difference in means is striking: thelandlocked countries outside of Western and Central Europe have an average income of $1,771,compared with the non-European coastal countries, which have an average income of $5,567 (p =.001). The difference in economic density is even stronger, since the landlocked countries tend tobe very sparsely populated (59 people per km2 in landlocked countries compared with 207 peopleper km2 in coastal countries).

8 The estimated significance level to the t-test is precise only if the errors are normally distributed, the two groups arestatistically independent, and they are drawn from a random sample (not the whole population). The small estimated p valuenevertheless indicates that the means are far apart. This caveat applies to the p values reported below.

9 There is, apparently, an inconsistency in our classification, since Tropics is based on land area in the geographicalzone, but Subtropics is defined according to ecozone (from Leemans 1990). We chose to rely on the geographical definition ofthe tropics both for convenience and because of its empirical relevance in the regression estimates. There is no comparablegeographical definition of the sub-tropics, so for that category we fall back on an ecozone definition.

10 The landlocked countries in Western and Central Europe are Austria, the Czech Republic, Hungary, the FormerYugoslav Republic of Macedonia, Slovakia, and Switzerland.

6

Of course, geography is not everything. Even geographically favored countries, such astemperate-zone, coastal North Korea, or well-located Czechoslovakia, failed to thrive under asocialist economic and political system. Nonetheless, development surely seems to be favoredamong the temperate-zone economies, especially the subset that: (1) is in the NorthernHemisphere; (2) has avoided socialism; and (3) has avoided being ravaged by war. In total, thereare 78 non-tropical economies, of which 7 are in the Southern Hemisphere: Argentina, Australia,Chile, Lesotho, New Zealand, South Africa, and Uruguay (all in the temperate zone). We classify46 countries as socialist during the post-war period, of which 31 are in the North temperate zone,and four are in the North sub-tropical zone. We also classify 12 non-tropical countries are war-torn. There are 12 non-tropical landlocked countries outside of Europe, of which 10 weresocialist and 2 were not.

With these definitions, we find the following. There are 23 countries with the mostfavored combination of geography and politics — Northern Hemisphere, temperate zone, coastal,non-socialist, and non-war torn — with an average income of $18,000. In fact, 22 of thesecountries have an average GDP per capita above $10,000, with Turkey and Morocco being theonly exceptions.11 Using a multiple regression estimate for the 78 non-tropical countries, averageincomes per capita are reduced by an estimated $4,785 for being in the sub-tropics; $3,590 forbeing in the Southern hemisphere; $10,053 for being socialist; and $5,190 for being landlocked.

A summary of geographical characteristics by major region is shown in Table 1. For eachregion, we show the average GDP per capita, total population and land area, and several keyvariables that we will find to be closely related to economic development: the extent of land in thegeographical tropics, the proportion of the region’s population within 100 km of the coastline orwithin 100 kilometers of the coastline or ocean-navigable river,12 the percentage of the populationthat lives in landlocked countries, the average distance by air (weighted by country populations) ofthe countries within the region to the closest “core” economic areas, 13 the density of humansettlement (population per km2) in the coastal region (within 100 km of the coastline) and theinterior (beyond 100 km from the coastline). Some important characteristics are the following. First, Sub-Saharan Africa, the poorest region, has several characteristics closely associated withlow income in general: a very high concentration of land in the tropics, a population heavily

11 Morocco is just on the borderline of classification as a sub-tropical country, with 48 percent of the population in the

tropical and sub-tropical ecozones.12 Using geographical information system (GIS) data, we calculate the coastal population in two ways. First, we take all

land area within 100 kilometers of the open sea, except for coastline in the arctic and sub-arctic region above the winter extent ofsea ice (NGS, 1995), and measure the population within that area. Additionally, we identify river systems that accommodateocean-going vessels on a regular basis (e.g. the Saint Lawrence Seaway and the Mississippi River), and add land areas within 100kilometers of such navigable rivers. Since we do not have direct costs of transport for these river systems, there are inevitablydifficulties in classification, as some ostensibly ocean-navigable rivers impose very high costs for such transport. Due to theclassification difficulties and the greater empirical relevance of coastal access alone, the population and land area near navigablerivers are not used in the rest of the paper. Information on our classification of river systems is available from the authors. Ocean-navigable rivers are displayed in Figure 8.

13We experimented with a number of distance measures, all of which produced similar outcomes. We therefore choosethe simplest: the smallest distance of the country’s capital city to one of the following three cities: New York, Rotterdam, orTokyo.

7

concentrated in the interior (only 19 percent within 100km of the coast); with more than a quarterof the population in landlocked countries (the highest of any region); very far from the closest“core” markets in Europe; and with low population densities in the coastal and interior regions. By contrast, Europe, the richest region shown, is non-tropical; heavily concentrated near coastalareas; with almost no population in landlocked regions; and with moderate population density. South Asia and the transition economies (of Eastern Europe and the former Soviet Union) are, likeSub-Saharan Africa, also heavily concentrated in the interior rather than the coast. India’s greatmass of population, for example, lives in the Gangetic valley, often hundreds of kilometers fromthe coast. South Asia is, of course, partly tropical and densely populated (indeed the most denselypopulated in the world), while the transition economies are non-tropical and also the least denselypopulated region. Latin America is the other highly tropical region, with low population densities,and with a moderately coastal population. The United States, not shown, has two enormousadvantages for development: a relatively high proportion of population near the coast (38 percentwithin 100 km of the coast, and a remarkable 67 percent if we include ocean-navigable riversystems), and a temperate-zone landmass.

These patterns prompt the following questions. How much has geography mattered foreconomic growth, once we control for economic policies and institutions? If geography matteredin the past, how much does it still matter today? Are there persistent advantages to earlydevelopers through agglomeration effects, learning by doing, and the like, or do latecomers havethe advantage of the possibility of rapid growth through technological diffusion, capital imports,and other forces of convergence?

Based on the evidence in this paper, we believe that geography continues to matterimportantly for economic development, alongside the importance of economic and politicalinstitutions. From an analytical point of view, we believe that geographical considerations shouldbe re-introduced into the econometric and theoretical studies of cross-country economic growth,which so far have almost completely neglected geographical themes.14 Our broad conclusions,described below, may be summarized as follows.

• Tropical regions are hindered in development relative to temperate regions, probably becauseof higher disease burdens and limitations on agricultural productivity

• Coastal regions, and regions linked to coasts by ocean-navigable waterways, are stronglyfavored in development relative to the hinterlands.

• Landlocked economies may be particularly disadvantaged by their lack of access to the sea,even when they are no farther than the interior parts of coastal economies, for at least threereasons: (1) cross-border migration of labor is more difficult than internal migration; (2)infrastructure development across national borders is much more difficult to arrange thatsimilar investments within a country; and (3) coastal economies may have military or

14The key recent reference on cross-country growth is Barro and Sala-i-Martin (1995). This book does not make a

single reference to economic geography. Nor do literally hundreds of recent cross-country growth studies. One recent (andnearly lone) exception is Hall and Jones (1996), who note that economic productivity of countries (measured as per capita GDP)increases with the distance from the equator.

8

economic incentives to impose costs on interior landlocked economies;

• High population density seems to be favorable for economic development in coastal regionswith good access to internal, regional and international trade. This may be the result ofincreasing returns to scale in infrastructure networks, or because of an enhanced division oflabor in settings of high population density. On the other hand, population density in thehinterland shows no such advantages, and our results show a net disadvantage.

• Population growth across countries in the recent past is strongly negatively correlated withtheir relative potential for economic growth. That is, human populations are growing mostrapidly in countries least equipped to experience rapid economic growth. More generally,there is no strong relationship between current population densities and a region’s potential formodern economic growth, since population densities seem more to have been driven byagricultural productivity rather than conditions for modern industry and services.

It is worthwhile to mention briefly the relationship of our approach to the recent creativeand important work on geography by Paul Krugman, Anthony Venables, and others. The “newgeography” follows the “new trade theory” by showing how increasing returns to scale,agglomeration economies, transport costs, and product differentiation can lead to a highlydifferentiated spatial organization of economic activity (including cities, hubs and spokes,international division of labor between industry and agriculture, etc.), even when the underlyingphysical geography is undifferentiated. These models illustrate the possibility of “self-organizing” spatial patterns of production based on agglomeration effects, rather than differencesin climate, transport costs, ecology, etc. Our starting point, by contrast, is that the physicalgeography is in fact highly differentiated, and that these differences have a large effect oneconomic development. The two approaches can of course be complementary: a city mightoriginally emerge because of cost advantages arising from differentiated geography, but thencontinue to thrive as a result of agglomeration economies even when the cost advantages havedisappeared. Empirical work should aim to disentangle the forces of differential geography andself-organizing agglomeration economies.

The paper is organized as follows. Section II discusses a range of theoretical approacheslinking geography and growth. Section III explores these linkages in cross-country empiricalmodels of growth, population density, and income per capita. Section IV looks in more detail atpopulation densities. Section V discusses several implications of these findings for economicpolicy and for future economic analysis.

II. Geography and models of economic growth

While geography has been much neglected in the past decade of formal econometricstudies of cross-country performance, economists have long noted the crucial role of geographicalfactors. Indeed, though Adam Smith (1776[1976]) is most remembered for his stress on economic

9

institutions, Smith also gave deep attention to the geographic correlates of growth.15 (Smith shouldalso be remembered for his recognition that Europe’s first-mover military advantage gave it anability to impose huge costs on other parts of the world. 16) Smith saw geography as the crucialaccompaniment of economic institutions in determining the division of labor. Smith’s logic, ofcourse, started with the notion that productivity depends on specialization, and that specializationdepends on the extent of the market. The extent of the market in turn depends both on the freedomof markets as well as the costs of transport. And geography is crucial in transport costs:

As by means of water-carriage a more extensive market is opened to every sort of industrythan what land-carriage alone can afford it, so it is upon sea-coast, and along the banks ofnavigable rivers, that industry of every kind naturally begins to subdivide and improveitself, and it is frequently not till a long time after that those improvements extendthemselves to the inland part of the country. (p. 25)

In view of the crucial role of transport costs, Smith notes that:

All the inland parts of Africa, and that part of Asia which lies any considerable way northof the Euxine [Black] and Caspian seas, the antient Sycthia, the modern Tartary andSiberia, seem in all ages of the world to have been in the same barbarous and uncivilizedstate in which we find them at present. The sea of Tartary is the frozen ocean which admitsof no navigation, and though some of the greatest rivers in the world run through thatcountry, they are at too great a distance from one another to carry commerce andcommunication through the greater part of it. There are in Africa none of those great inlets,such as the Baltic and Adriatic seas in Europe, the Mediterranean and Euxine seas in bothEurope and Asia, and the gulphs of Arabia, Persia, India, Bengal, and Siam, in Asia, tocarry maritime commerce into the interior parts of that great continent . . . (p. 25)

Great thinkers such as Fernand Braudel (1972, 1981-84) and William McNeill (1963,1974), and important recent historians such as E. L. Jones (1981) and Alfred Crosby (1986), haveplaced the geography and climate of Europe at the center of their explanations for Europe’s pre-eminent success in economic development. Braudel pointed to the key role of the Mediterranean-based and North-Atlantic coastal economies as the creative centers of global capitalism after thefifteenth century. McNeill similarly stressed Europe’s great advantages in coastal trade, navigablerivers, temperate climate, and suitable disease patterns as fundamental conditions for Europeantakeoff and eventual domination of the Americas and Australasia. Crosby details the advantages of

15Smith does not discuss culture and economic development in any detail in the Wealth of Nations, but it seems clear

that Smith, in line with much thinking of the Scottish Enlightment, viewed human nature as universal, and did not view culture as aprimary differentiating factor in economic development. After all, Smith saw the “propensity to truck, barter, and exchange onething for another,” to be universal, not culturally specific. For example, Smith never bemoans the lack of entrepreneurial zeal inone place or another as an explanation for poor economic performance. Later thinkers such as Max Weber put great stress onculture, though the alleged linkages are particularly difficult to document and test. Recently, one of the leading economic historiansDavid Landes (1998) has argued that culture, in addition to geography and institutions, should be given pride of place in explainingthe differences in economic performance.

16He noted that “all the commercial benefits” to the East and West Indies that might have resulted from increased trade“have been sunk and lost in the dreadful misfortunes” occasioned by European military advantage, which enabled the Europeans“to commit with impunity every sort of injustice in those remote countries.”

10

the temperate zones in climate, disease ecology, and agricultural productivity. Two importantessays, one by the Council on Foreign Relations (Lee, 1957), and one a generation later byAndrew Kamarck (1976) for the World Bank, synthesized these arguments in excellent surveys onTropical Development, which have been largely ignored by the formal modelers of economicgrowth.

One of the most interesting recent attempts to ground very long-term development in

geographical and ecological considerations comes from ecologist Jared Diamond (1997), whoasks why Eurasians (and peoples of Eurasian origin in the Americas and Australasia) “dominatethe modern world in wealth and development” (p. 15). He disposes of racialist explanations notjust on moral grounds but on rigorous findings of the shared genetic inheritance of all humansocieties. His explanation rests instead on the long-term advantages of Eurasia in agglomerationeconomies and the diffusion of technologies. Human populations in the Americas and Australasiawere cut off by oceans from the vast majority of human populations in Eurasia and Africa. Theytherefore could not share, through trade and diffusion, in technological advances in agriculture,communications, transport, and the like. Additionally, Diamond argues that technological diffusionnaturally works most effectively within ecological zones, and therefore in an East-West directionalong a common latitude, rather than in a North-South direction, which almost invariably crossesecological zones. This is because plant species and domesticated animals appropriate to oneecological zone may be completely inappropriate elsewhere. Eurasia, claims Diamond, thereforeenjoyed the benefit of its vast East-West axis heavily situated in temperate ecological zones, whileAfrica was disadvantaged by its North-South axis which cut across the Mediterranean climate inthe far North, the Saharan Desert, the equatorial tropics, and the Southernmost sub-tropicalregions. Diamond argues that these advantages, in addition to more contingent (i.e. accidental)advantages in indigenous plant and animal species, gave Eurasia a fundamental long-termadvantage over the rest of the world.

Historians have also stressed the changing nature of geographical advantage over time, astechnology changes. In early civilizations, when transport and communications were too costly tosupport much inter-regional and inter-national trade (and virtually any oceanic trade),geographical advantage came overwhelmingly from agricultural productivity rather than fromaccess to markets. Therefore, early civilizations almost invariably emerged in highly fertile rivervalleys such as the Nile, Indus, Tigris, Euphrates, Yellow and Yangtze rivers. These civilizationsproduced high-density populations that in later eras were actually disadvantaged by theirremoteness from international trade. Northern Europe could not be densely settled before thediscoveries of appropriate technologies (e.g. the moldboard plow in the Middle Ages) and tools tofell the great Northern forests (Landes, 1998). Similarly, as the advantages of over-land trade andcoastal-based trade between Europe and Asia gave way to oceanic commerce in the 16th century,economic advantage shifted from the Middle East and Eastern Mediterranean, to the NorthAtlantic. In the 19th century, the high costs of transport of coal for steam power meant that earlyindustrialization almost invariably depended on proximity to coal fields. This advantage of coursedisappeared with the discoveries of petroleum refining, oil and hydro-based electricityproduction, and reduced cost of bulk transport. Railroads, automobiles, and air transport, as wellas all forms of telecommunications, surely reduced the advantages of the coastline relative to thehinterland, but according to the evidence below, the advantages of sea-based trade remain.

11

To summarize, we can say that leading historians and economists have long recognizedgeography as a crucial scaffolding for economic development, even though geography has beenneglected in most recent empirical studies of comparative growth. Leading thinkers have pointedto four major areas in which geography will play a fundamental direct role in economicproductivity: transport costs, human health, agricultural productivity (including animal husbandry);and proximity and ownership of natural resources (including water, minerals, hydrocarbondeposits, etc.). The factors may also have indirect effects, if first-mover advantages or populationdensities affect subsequent growth dynamics through agglomeration economies or other feedbackmechanisms. We now turn to a more formal consideration of these factors.

Formal Models of Geography and Development

To establish some formal ideas about the interaction of geography and development, westart with the simplest model of economic growth, the AK model (known in its earlier incarnationas the Harrod-Domar model), and add transport costs. In the resulting model, growth differencesacross countries depend on several parameters that we will find to be important in later empiricalanalysis. These factors include: (1) underlying total factor productivity, denoted by A, which maydiffer across countries for fundamental geographical reasons (e.g. the differences in productivity oftemperate and tropical agriculture; differences in endemic health conditions among variousecozones); (2) transport costs, reflecting both distances and physical access to trade (e.g.navigability of rivers, distance from the coastline); and (3) national saving rates, and implicitly,government economic policies.

Suppose that an economy has the aggregate production function:

Q = AK

The capital stock evolves according to:

dK /dt = I - δK

We assume for the moment that population is constant, and is normalized to 1, so that Q representsboth output and output per capita. Later on we discuss some issues of population growth.

The national saving rate is fixed at s (the alternative assumption of intertemporaloptimization would be straightforward as well, but with little gain in realism or insight). Therelative price of investment goods to final output is PI. Thus,

sQ = PII

The growth rate of the economy is then

12

(1) γ = (1/Q)(dQ/dt) = (1/K)(dK/dt) = sA/PI - δ

Economic growth is positively dependent on the saving rate, s, and the level of productivity, A, andnegatively dependent on the relative price of capital goods, PI, and the rate of depreciation, δ.

Transport costs affect the relative price of capital goods because some investment goodsmust be imported from abroad. In many developing countries, virtually all equipment investmentis imported from abroad. To illustrate some implications of transport costs, we now assume thateach country produces a distinct final good, and that investment I is a composite of the final goodsproduced in the various countries. The key assumption is that there are gains from trade, so thattransport costs and other barriers to trade reduce growth. We do not directly model the underlyingreasons for specialization in production, and hence gains from trade. As is well known,specialization in production may result from some or more of the following factors: differingprimary factor endowments; economies of scale in production; economies of specialization vialearning by doing; or differing technologies across countries because of investments in proprietaryR&D.

Total investment I depends on investment expenditure on domestic goods Id and onimported investment goods Im:

I = I(Id , Im )

As a simple illustration, consider the case in which I is a Cobb-Douglas function of the underlyingdomestic and foreign investment goods:

I = (Id)a(Im)(1-a)

The true price index of investment goods, then, is a geometric average of the price of domesticinvestment goods and foreign investment goods. Setting the domestic good as numeraire, i.e. thethe price of Id equal to 1, we have:

(2) PI = α(Pm)(1-a) where α = aa (1-a)(1-a)

We denote the (exogenous) world market price of the imported good as Pm*, and write thelanded (or cif) price in the home economy as Pm = τPm*, where τ > 1 is the cif factor, i.e. theworld price plus cost, insurance, and freight. Then, from equations 1 and 2 we have a modifiedequation for the growth of the economy:17

17Suppose that instead of Cobb-Douglas, the investment function is a constant-elasticity-of-substitution (CES) function

of the underlying investment in each country: I Ii i= ∑ − −

µ ε ε1

. The elasticity of substitution is σ ε= +1 1( ) . In that case,

the price index PI is also a CES of the prices of the individual investment goods, of the form: P PI j j j= ∑ − −

µ τσ σ σ

( )* ( ) ( )1 1 1

.

Thus, if the investment function has an elasticity of substitution σ, the price index has an elasticity of substitution 1/σ. When σequals 1, so that the investment function is Cobb-Douglas, the price index is also Cobb-Douglas, with

P vP v P v a a aI j

a

j

a

j

a a a

n

aj j j n= ∏ = ∏ =( ) ,* τ 1 21 2 K . The growth rate of the home country depends on a geometric weighted

13

(3) γ = (sA/α)(Pm*)-(1-a)τ--(1-a) - δ

The growth rate is now inversely related to the cost of transport, τ. Transport costs in this modelreduce growth by raising the cost of the imported capital good, thereby lowering the growth rate.We have seen in earlier empirical studies of economic growth (Barro, 1991, for example) that therate of growth is a decreasing function of the relative cost of investment goods. This is essentiallythe channel by which the costs of transport and distance enter in equation 3.

Equation 3 suggests three important points even at this very basic and abstract level. First,growth rates will differ according to underlying total factor productivity A. Second, growth ratesdiffer according to transport costs τ. These, in turn, are likely to depend on severalcharacteristics. Coastal economies will generally have much lower transport costs than hinterlandeconomies. Countries near to core economies (e.g. the main capital-goods providers) willgenerally have lower transport costs than distant economies, so that growth is likely to diminish indirect relation to the distance from the core. Third, protectionist policies that raise the domesticprice of imported capital goods, or that limit the exports needed to import the foreign capitalgoods, are likely to reduce long-term economic growth. We can’t overstress the practical pointenough: countries require capital goods imports for long-term growth. Protectionist policies raisethe price of those imports and thereby slow growth.

A model with intermediate goods

Suppose now that final production requires imported intermediate inputs. This assumptionis of enormous empirical importance, since many of the key manufactured exports of developingcountries involve the importation of intermediate manufactured goods (e.g. fabrics, electroniccomponents), which are then assembled domestically with low-cost labor and re-exported toworld markets. The transport costs involved in the import of intermediate products, and their re-export after domestic processing can be of critical importance in the success or failure of themanufacturing export sector, even if the transport costs for investment goods are minimal.

We must now distinguish between gross output Q and gross domestic product, Y. Inparticular we set:

average of transport costs from each of its capital suppliers, with the weight equal to the share of investment goods from j in thetotal investment expenditure of the home country. When the elasticity of substitution among investment goods is infinite, the CES

price index has zero elasticity, and takes the form: P PI j j j= min *τ µ . What counts in that case is not a weighted average of

prices, but rather the lowest price adjusted for the productivity of the various investment goods. The relevance for geography

would be as follows: assuming that the efficiency-adjusted price of capital goods is equal in all markets, i.e. Pj j

* µ is the same

across all markets, growth would depend on the minimum distance to one of the capital-goods suppliers, rather than to theaverage distance to all of the suppliers of capital goods. In practice, transport costs would depend on the minimum distance to amajor market — the U.S., or Western Europe, or Japan — rather than the average distance to the U.S., Europe, and Japan. Inour empirical work, we find that the specification of minimum distance to a major capital goods supplier outperforms the averagedistance to all major capital goods suppliers.

14

Q = min[AK, N/µ]

where N is the intermediate good imported from abroad. The final good in the home marketcontinue to be numeraire (with price 1), and the relative price of the imported intermediate good isPn = τPn

*. The gross domestic product in units of the final good is given by: Y = AK - PnN = AK -µPnAK, or:

(4) Y = (1 - µPn)AK

Since domestic final output in the home market is numeraire, its price in the foreign marketinclusive of transport costs is τ. Similarly, if the intermediate product sells for Pn in the homemarket, its price in the foreign market is Pn/τ. Suppose that at world prices (i.e. in the foreignmarket) the share of the intermediate good in final output is given by σ = (Pn/τ)N/(τQ). Then,equation 4 may be re-written as:

(4') Y = [1 - στ2]AK

All of the model goes through as before, so that the modified growth equation is now:

(5) γ = (sA/α) [1 - στ2] (Pm )-(1-a) - δ

The key point here is that relatively small transport costs can have huge effects on outputand growth when the share of intermediate inputs in final demand is large. For example, supposeσ = 0.7. Now, compare the growth rates of countries with one-way transport costs equal to 5percent and 10 percent, i.e. τ = 1.05 versus 1.10. Ignore, for the moment, the transport costs oncapital goods, to focus solely on the effect of intermediate products. Let γ1 be the growth rate ofthe low-transport-cost economy, and γ2 be the growth rate of the high-transport-cost economy. Then, from equation 5:

γ1 / γ2 = [1-.7(1.1025)]/[1-.7(1.21)] = 1.49

The growth rate in the low-transport-cost economy is 49 percent higher than in the high- transport-cost economy, even though the transport costs differ by only 5 percentage points. The explanation,of course, is that a “mere” 5 percentage point decrease in one-way transport costs on intermediateand final goods implies a whopping increase of 49 percent in domestic value added.

The notion that intermediate inputs represent such a high proportion of the value of grossoutput may seem unrealistic, but such is the case for many key export sectors in developingcountries. In many kinds of labor-intensive industries — such as apparel and electronics assembly— the developing country imports a very high proportion of the value of final output. The

15

intermediate imports are assembled by the domestic workers, and then typically re-exported toworld markets. The developing country is essentially selling labor services used in assemblyoperations, rather than selling the entire product. For such assembly industries, even smallincreases in transport costs can render the sector non-competitive. Thus, in Radelet and Sachs(1998) we find that only those developing countries with good transport access to world marketshave been able to establish assembly-type industries.

These arguments further underscore the disadvantages of the hinterland relative to the coastin economic development. Almost all modern production depends on the multi-stage processing ofoutput, with inputs often produced in many specialized enterprises, some abroad and somedomestic. The low-cost transport of such intermediate products is crucial, especially so indeveloping countries where many intermediate components are necessarily imported from abroad. Only coastal areas, or areas linked to the coastline through navigable waterways or very low-costland transport, have a chance to compete in such activities.

Divergence, Convergence, and Poverty Traps

The AK model, of course, has one central feature: the absence of convergence. Becausethere are no diminishing returns to investment in the production function, there is no tendency forgrowth to slow down as capital deepening occurs. For this reason, countries that have underlyingadvantages in saving rates, efficiency, transport costs, or rates of depreciation, will displaypermanently higher growth rates and a widening proportionate gap with slower growing countries.

Models like the AK model also highlight another possibility. Suppose that transport costsare important at one stage of history in determining economic location, but then become lessimportant. In the AK model, the early advantage would lead to a boost of economic activity so thatthe early-favored region would jump ahead of the others. Once the advantage is lost, alleconomies would grow at the same rate (assuming the same A, s, δ). The early advantage wouldnever be lost in terms of relative income levels, though growth rates would converge. In modelswith increasing returns to scale rather than the constant returns to scale in the AK model, the earlyadvantage could lead to persistently higher growth rates even if the geographical advantage itselfdisappears, since growth rates could be a positive function of the level of capital, which itselfwould be raised by the transitory advantage. Thus, one possible interpretation of the introductoryobservations that temperate, coastal economies have the highest levels of GDP is that suchgeographical attributes once conferred advantages in the past, even if they no longer do so today.

As is well known, if the foregoing model is recast as a neoclassical growth model withdiminishing marginal productivity of capital, so that Q = AKβ , with β < 1, then the conclusionsreached so far have to be re-cast as follows. The same list of parameters (s, A, Pm*, τ, δ) now allaffect the steady-state level of Q, denoted Qss, and the steady-state capital stock, Kss, but not thelong-term growth rate. Because the capital stock converges gradually to its steady state, so too

16

does the level of output. In this case, we can write the growth equation as follows:

(6) γi = (1/Qi)(dQi /dt) = λ[lnQssi - lnQi ]

Equation 6 holds that the proportionate rate of growth of country i, γ i , depends on the gap betweenthe steady-state level of output and the contemporaneous level of output. We could, in general,derive ln Qi

SS to be a function of the underlying parameters s, A, etc. Approximating thisrelationship in log-linear form, ln Q Zi

SSi = ′β , where Zi is the vector of underlying growth-

influencing parameters, and β is a vector of coefficients, we end up with an empirically estimableequation that has been extremely popular in recent years:

γ λβ λi i iZ Q= ′ − ln

In this formulation, growth depends positively on the parameters in question, and negatively on theinitial income variable. The empirical presence of the term λlnQi has been used to examinewhether or not there is a tendency towards convergence as in the neoclassical model, or continuingdivergence, as in the AK model in which the level of income is not a determinant of the rate ofgrowth.

As is well understood, the issue of convergence versus non-convergence depends heavilyon the underlying structure of production. In an environment of increasing returns to scale at thefirm level as well as gains from a greater diversity of products — as in the popular models ofDixit-Stiglitz imperfect competition (Romer 1986, 1990; Grossman and Helpman, 1991) — theremay well be increasing or at least constant returns to the capital stock at the macroeconomic level. The marginal productivity of capital would then be constant (as in the AK model) or evenincreasing, as the aggregate capital stock increases. In that setting, the AK model would depict theaggregate production technology better than the neoclassical model with its assumption ofdeclining marginal productivity of capital. Thus, as is well known, to the extent that scaleeconomies and product diversity are critical, we would expect to see little convergence betweenrich and poor countries, and could well see divergence. To the extent that economies of scale andproduct diversity are limited, we would be more likely to see convergence in income levels,controlling for other factors.

Geography and population dynamics

It is not easy to integrate population dynamics in a meaningful way into the simple AKmodel, so we’ll have to step outside that simple framework to discuss some important issues onthe relationship of geography, population dynamics, and growth. We have stressed repeatedly theadvantages of coastal regions for economic development. We have not said anything, however,about the distribution of human populations across regions. In fact, the linkages are problematic inthe following three senses. First, there are vast human populations in regions quite disadvantagedfor modern economic growth. In the long course of human history there has been one tendency for

17

human population densities to rise in areas favorable for growth, so that coastal areas are indeedmore densely populated than hinterlands. Another force, however, has been for populationdensities to rise in fertile agricultural areas, for example along inland river systems like theGanges, Tigris, Euphrates and Nile, that are useful for irrigation and inland trade, but notinternational trade. The result is high population densities living in subsistence agriculture ratherbenefiting from modern economic growth. Second, current population growth tends to be highestin the more remote regions, mainly because population growth is negatively related to per capitaincome, and especially inversely related to mother’s education and the market value of mothers’time.18 Thus, the concentration of populations in problematic regions is growing. This effect isstrengthened by the tendency for improvements in public health (e.g. the spread of antibiotics,vaccines, oral rehydration therapy, etc.) to diffuse more readily than economic growth itself, sothat population growth has risen sharply since 1950 in some of the poorest regions.

Third, as a result of the mismatch of economic growth and population trends, there is amass migration of populations from the hinterland to the coast. The vast majority of migratorymovements are within poor countries, leading to unprecedented inflows of population into urbanareas and the rise of mega-cities in developing countries. The next largest migration, most likely,is across borders of developing countries, including vast flows of population from landlockedcountries to coastal economies. And the third largest migration is from poor countries to richercountries. This migration would of course be vastly larger were it not for immigration controls inthe richer economies. In any case, the pressures for migration both internal and international willrise sharply in the coming decades as differences in income levels continues to increase.

The effects of population pressures on economic growth are likely to differ markedly in thehinterland and the coast. In the hinterland, where transport costs are extremely high, the divisionof labor is low, and output is most likely characterized by decreasing returns to scale in labor inthe face of limited supplies of land. Therefore, higher population densities will be associated withfalling output per capita, a tendency that we have seen in many African countries in the past 20years. In the coastal economies, on the other hand, where transport costs are low and the divisionof labor is high, a rising population may be associated with stable or even increasing incomes percapita, even when the capital-labor ratio declines. This is because higher population densitiesmake possible an increasingly refined division of labor.19

18Caldwell (1982), in particular, has argued that population pressures are likely to remain high in rural areas while falling

sharply in urban areas. According to Caldwell, children are net economic assets in peasant rural areas since they can assist inhousehold production from an early age (e.g. carrying water and firewood), do not generally require high expenditures oneducation, and can be counted on to care for parents in old age. In an urban setting, however, children are net economic costs:they are likely to attend school rather than contribute to household production, and because of urban mobility, are much lessreliable as social security for aged parents. Moreover, the opportunity costs of raising children are much higher, especially ifwomen are part of the urban labor force.

19The standard model of traditional agriculture is based on a neoclassical production function of the form Q = Q(L,F),where L is labor input and F is a vector of other farm inputs, including land. Output per capita falls as L rises relative to F. Forthe modern sector, the increasingly popular model of differentiated production makes Q a function of n intermediate products Xi,

each produced with labor, Li, so that Q X i= ∑ γ γ( )1

, with 0 < γ < 1. Under conditions of monopolistic competition, free

entry, and costless introduction of new product varieties, it is typically shown that Xi is fixed by profit maximizing firms at a givenproduction run x, with Li = a + bx. For a total labor force L Li= ∑ , we have n(a+bx) = L, or n = L/(a+bx). Then,

18

We thus see that economies are likely to bifurcate on two pathways. The hinterland will becharacterized by decreasing returns to scale in labor, and high rates of population growth. Thecoastline will be characterized by increasing returns to scale, and falling rates of populationgrowth as incomes per household rise. The hinterland may therefore show Malthusian dynamicswhile the coastline will show rising income levels and falling natural population growth rates. The two systems will interact through ever-greater pressures on migration from the hinterland tothe coast.

Geography and policy choices

So far we have stressed that geography may influence growth directly through the level ofproductivity and transport costs. Geography can have another potent effect by affecting the choiceof economic policies. Countries that are proximate to markets, for example, may choose moreopen trade policies than countries that are distant from markets. We offer a motivation for such apossibility.

Suppose that growth is given by sA/PI , and that A itself can be decomposed into amultiplicative policy component, π, and a purely exogenous component, θ: sπθ/PI. Suppose aswell that PI = PI(τ), that is the price of investment goods is an increasing function of transportcosts. At this abstract level, we can say that the policy component of growth is a decreasingfunction of the ad valorem tax rate levied by the government on the private economy π = π(Τ). Forsimplicity, we will use a linear functional form: π = c - eT. These taxes might be formal taxes,bribes demanded to clear customs, seizures of property, etc. The basic idea is that the governmentgains revenues but at the expense of a worsening policy environment.

Suppose that T is chosen once and for all to maximize the expected utility of governmentofficials. To keep matters very simple in this abstract framework, we assume that the policymaker has an intertemporal log utility function, a pure discount rate of d, and a hazard rate h oflosing office.20 Expected utility is then:

(7) EU e Q t dtd h t = -( + )z log[ ( )]Τ

This is maximized subject to the constraint that Q t Q e t( ) = 0γ where γ = sπ(T)θ/PI(τ). What we

have, essentially, is an “optimal tax” calculation on the part of the sovereign. Higher taxationyields more immediate revenue, but at the cost of slower future growth, and hence lower futurerevenues. Simple manipulations show that equation 7 can be re-written as

production is ∑ = = + −x n x L x a bxγ γ γ γ γ( ) ( ) ( ) ( )( )

1 1 1 1 . The result is that production shows increasing returns to labor.

Output per labor is therefore a rising function of L. 20There is an instantaneous probability h of losing office. The probability distribution for tenure in office is then

f t he ht( ) = - , and the mean time in office is 1/h.

19

EU = [log(T)+log(Q0)]/(d+h) + [sπ(T)θ/PI(τ)]/(d+h)2 . It then remains to calculate the optimal T,by setting dEU/dT = 0. We find:

(19 ) T = PI(τ)(d+h)/sθe T ≤ 1

This is an insightful expression. The optimal tax is an increasing function of transportcosts, discount rate, and probability of losing office; and a decreasing function of total factorproductivity and the responsiveness of growth to the tax rate. Basically, the sovereign is tradingquick gain for future loss. To the extent that growth is low (because PI is high or θ is low), or thefuture is heavily discounted (because d+h is high), or is unresponsive to taxation, then the tax rateshould be set at a high level. To the extent that underlying growth is rapid or is highly responsiveto taxation, then T should be set at a lower rate.

The implications for geography are as follows. First, good policy and good geographymay have a tendency to go together. When growth is inherently low because of adversegeographical factors, and also unresponsive to policy (perhaps for the same reasons), the revenue-maximizing sovereign will impose high rates of taxation, e.g. protectionist policies. When theeconomy is inherently productive and responsive to good economic policies, the sovereign willhave the incentive to impose low rates of taxation. The result is that natural differences ingrowth potential tend to be amplified by the choice of economic policies.

Second, the correlation of favorable underlying growth conditions and good policies leadsto an important identification problem in estimating the effects of economic policy on economicperformance. Suppose that one carries out a regression estimation of growth on taxation, and findsa strong negative relationship. This is usually interpreted as a demonstration that policy mattersfor growth. We have just seen, however, that it might also reflect the fact that growth matters forpolicy. It is crucial to specify structural growth relationships that include both policy andunderlying geography in order to disentangle these alternatives.

III. Empirical linkages of geography and growth

Geographical correlates of growth

The basic theory points to two broad categories of geographical factors of deepsignificance: transport costs, measured by the parameter τ , and intrinsic productivity, measured byA. Consider first the transport costs. Remarkably, despite the likely importance of transport costsfor economic growth, there are no adequate measures of transport costs for a large sample ofdeveloped and developing countries. The best that we could obtain for a large number ofcountries is the IMF estimates of the CIF/FOB margins in international trade. These marginsmeasure the ratio of import costs inclusive of insurance and freight (CIF) relative to import costsexclusive of insurance and freight (FOB). There are several problems with these measures, themost important being that: (1) they are only crudely estimated by the IMF staff; and (2) they dependon the composition of imports, and thus are not standardized across countries.

20

Nonetheless, the CIF/FOB margins are informative, and predictive of economic growth. The CIF/FOB margin for 1995 is 3.6 percent for the U.S., 4.9 percent for Western Europe, 9.8percent for East Asia, 10.6 percent for Latin America, and a whopping 19.5 percent for Sub-Saharan Africa.21 We estimate an equation relating the CIF/FOB band to the distance of thecountry to the “core” areas of the world economy (Distance, measured in thousands of kilometers),and to the accessibility of the country to sea-based trade, by including a dummy variable for non-European landlocked countries (Landlocked):

CIF/FOB = 1. 06 + 0.010 Distance (1,000 km) + 0.11 Landlocked (84.9) (3.0) (2.4)

N = 83, R2 = 0.3222

As expected, there is a penalty both for distance from the core economies and for beinglandlocked. Each 1,000 km raises the CIF/FOB margin by 1.0 percentage points; being landlockedraises the CIF/FOB margin by 11.1 percentage points. We will show later that the CIF/FOBmargin is indeed predictive of income levels and economic growth. Incidentally, if we regressCIF/FOB on the proportion of the population within 100 km of the coast (Pop100km), we also finda negative effect, but when we include both Pop100km and Landlocked, Landlocked has moreexplanatory power. The coefficient on Pop100km remains negative, as expected, but drops inmagnitude and is statistically insignificant.

We have seen in Table 1 that the various regions in the world differ markedly in locationsof their populations relative to the seacoast and navigable rivers. African populations especiallyare far from the coastline, while Europe is overwhelmingly coastal. Indeed, Africa has the highestproportion of landlocked population of any continent in the world, and especially in East Africa,the populations are heavily in the interior (beyond 100 km of the coast) even in countries withcoastlines such as Kenya (coastal population equals 6 percent of the total), Mozambique (40percent), Sudan (2 percent), and Tanzania (16 percent). The situation is made worse by the factthat Africa’s interior regions are not accessible by ocean-navigable vessels, since the riversystems in Africa almost all face impassable barriers (e.g. cataracts, shallows, etc.) that preventthe entry of ocean-going vessels into the interior of the continent.

The notion that the coastal access has a large effect on trade and growth is plausible givenwhat we know about the growth patterns of the most successful developing countries in the periodafter 1965. Almost without exception, fast-growing developing countries have based their rapidgrowth on labor-intensive manufacturing exports. And almost without exception, such activitieshave expanded in port cities or export zones close to ports. As shown in Radelet and Sachs(1998), almost all countries with macroeconomic success in labor-intensive manufacturing exportshave populations almost totally within 100 km of the coast.

21The data refer to unweighted country averages for the respective regions for the countries for which IMF data are

available.22 Absolute value of t statistics are in parentheses below the coefficients, and N is the number of observations.

21

The geographical determinants of the efficiency parameter A are potentially much morevaried. First, we can surmise that coastal access will matter for internal trade and productivity aswell as for international trade. Cities are engines of growth, and most large cities other thangarrison towns or administrative capitals typically grow up on coastlines or ocean-navigablerivers. Therefore, countries with neither coastlines nor navigable rivers tend to have lessurbanization, and less growth. A simple regression estimate for 149 developed and developingcountries in 1995 shows that more ocean-accessible regions in the world are indeed also moreurbanized, as are economies closer to the economic core regions. In a simple regression we find:

%Urban = 132.3 + 17.1 Pop100km - 10.8 LDistance R2 = 0.29 (10.8) (3.6) (7.1) N = 149

A second major dimension of productivity linked to geography is the prevalence ofinfectious disease. As shown in Figure 4, malaria, with an estimated incidence of between 200and 500 million cases per year (WHO,1997a), is almost entirely concentrated in the tropics.23 Thispattern is neither accidental, nor mainly the result of reverse causation in which poor countries areunable to eradicate a disease under control in rich countries. There is no effective prophylaxis orvector control for malaria in the areas of high endemicity, especially Sub-Saharan Africa. Earliermethods of vector control are losing their effectiveness because of increased resistance of themosquitoes to insecticides. Standard treatments are also losing effectiveness because of thespread of resistance to chloroquine and other antimalarial drugs. The geographical extent ofmalaria is determined mostly by the ecology of the parasites (different species of malariaPlasmodia) and the vectors (different species of Anopheles mosquitoes). Malaria was broughtunder control since 1945 mainly in temperate-zone and sub-tropical environments, where thefoothold of the disease (both in terms of the mosquito population and the parasite endemicity) wasmore fragile. Figure 5 shows the extent of endemic malaria in 1946, 1966 and 1994, with itsgradually retreat to the core tropics. Using data from the World Health Organization, we constructa measure of malaria intensity (seen in Figure 4).24 A simple regression of malarial intensity onecological zones shows that it is most intense in the tropics and somewhat less in the subtropics(relative to all other ecozones). There is also a strong “Sub-Saharan Africa” effect.

23The great range of uncertainty about the number of cases is itself indicative of the lack of concerted study and

monitoring of malaria by international organizations in recent years.24Because of lack of true malaria incidence or prevalence data for the most severely affected countries, our index is

necessarily approximate. We digitized a world map of the extent of malaria in 1994 (shown in Figure 5) from the World HealthOrganization (WHO 1997), and used GIS to calculate the fraction of a country’s land area subject to malaria, excluding the areasof “limited risk”. To quantify the differing intensity of malaria, we collected WHO (1992) data for 1990 on the percent of malariacases that are the malignant falciparum species of malaria, which, of the four species of malaria, has the most severe symptoms,is the most resistant to drugs, and is responsible for almost all malaria mortality. The malaria index is the product of the percentland area times the percent of falciparum cases. As discussed below, falicparum malaria is predictive of low economic growth,but non-falicparum malaria is not.

22

Malarial Intensity (scale 0-1) =

-0.01 + 0.3 Wet Tropics + 0.5 Dry Tropics + 0.4 Wet Subtropics + 0.2 Dry Subtropics (0.7) (2.0) (4.9) (3.9) (1.5)

+ 0.5 Sub-Saharan Africa N = 148 R2 = 0.74 (6.7)

The pattern for malaria is common for a range of infectious diseases, whose vectors of

transmission depend on the tropical climate. Many diseases that are carried by mosquitoes(dengue, yellow fever, and lymphatic filariasis in addition to malaria), mollusks (e.g.schistosomiasis), and other arthropods (onchocerciasis, leishmaniasis, trypanosomiasis, Chagas’disease, and visceral filariasis), are endemic in the tropical ecological zones and nearly absentelsewhere. While data on disease burdens by country are generally not available, the recentmassive study by Murray and associates on the burden of disease (Murray and Lopez, 1996),confirms the heavy tropical concentration of infectious disease as a cause of death. This is shownin Figure 6.

A third major correlate of geography and productivity is the link of climate and agriculturaloutput. Our own estimates of agricultural productivity suggest a strong adverse effect of tropicalecozones on the market value of agricultural output, after controlling for inputs such as labor,tractors, fertilizer, irrigation and other inputs. Our estimates in Gallup (1998) suggest that tropicalagriculture suffers a productivity decrement of between 30 and 50 percent compared withtemperate-zone agriculture, after controlling as well as possible for factor inputs.

Other more prosaic geographical correlates, such as the endowments of high-value naturalresources (hydrocarbons, minerals, precious gems) also affect cross-country income per capita ata point in time. We do not have comprehensive measures of the international market value of suchresource endowments, so we settle for a rough measure of one key resource: deposits of petroleumand natural gas. Countries rich in hydrocarbon deposits per capita indeed display higher levels ofper capita income in 1995, though not necessarily higher economic growth. Indeed, Sachs andWarner (1995a) have suggested that higher natural resource exports as of 1970 (measured as apercent of GDP) are negatively related to subsequent growth.

Geography and levels of per capita income

We examine the linkage of output to geography both in levels and rates of change of GDP. Suppose that countries differ in their growth rates according to a vector of characteristics Z,including determinants of transport costs and total factor productivity. We write a linearapproximation of a growth equation like (6) such that:

(20) γ β µit i itZ= ′ +

23

In any period T > 0, QiT = exp(γi T)Qi0 . Suppose that in the distant past Qi0 is randomly distributedacross countries, with lnQi0 = lnQ0 + ζi , and with ζi independent of the Zi. Then, we can estimatea cross-country level equation of the form:25

(21) ln( ) ln( )Q Q T ZiT i i= + ′ +0 β ε

The effects of the parameters Zi on the level of QiT will tend to grow over time, in proportion to T,since Z affects the growth rate, not merely the relative level of income, of country i. If the Zvariables are time dependent, then QiT is a function of the entire time path of Zi. In general, for mostvariables of interest, we have snapshots of Z, rather than a time series. One objective of empiricaldevelopment studies should be the creation of time series for measures of key institutionaldeterminants of growth (e.g. openness of markets, protection of property rights, etc.) in order tostrengthen our empirical tests.

We start with the simplest specification, writing the log level of per capita income as afunction of three underlying geographical variables: (1) Tropicar, the percentage of land in thegeographical tropics; (2) Pop100km, the proportion of the population within 100km of thecoastline; and (3) LDistance, the minimum log-distance of the country to one of the three coreregions, measured specifically as the minimum log-distance to New York, Rotterdam, or Tokyo. We estimate this relationship three times: for Maddison’s GDP estimates for 1950 and 1990, andfor the World Bank’s PPP GDP estimates for 1995 on the subset of countries for whichMaddison’s data are available. In all three regression estimates, reported in Table 2 output is apositive function of Pop100km, and a negative function of Tropicar and LDistance. The magnitudeof the effects tends to increase over time, as expected. In 1950, the “penalty” for Tropicar was -0.69, signifying that tropical areas were only 50 percent (= exp(-0.69)) of per capita income of thenon-tropical areas controlling for the other factors. By 1995, the effect has risen to -0.99 (or 37percent of the non-tropical areas). Similarly, the benefit of a coastal population rose from 0.73 in1950 to 1.17 in 1995. The suggestion is that being tropical, landlocked and distant was badalready in 1950, and adverse for growth between 1950 and 1995.

We now turn in more detail to the 1995 data, for which we have a wider range of possibleexplanatory variables. We start by estimating this simple level equation for GDP per capita on aPPP-basis in 1995, for the 150 countries with population greater than 1 million, and then turn togrowth equations for the period 1965 - 90. We group the explanatory variables Z into three broadcategories: (1) variables related to transport costs and proximity to markets; (2) variables relatedto ecological zone; and (3) variables related to economic and political institutions.

In regression (4) of Table 2, we limit Z to a parsimonious set of four variables closelylinked to geography: the prevalence of malaria; transport costs as measured by the CIF/FOBmargin; the proportion of the country’s population near the coastline; and the endowment ofhydrocarbons per capita. These four variables alone account for 69 percent of the cross-countryvariation in per capita income, and are all with the expected sign and statistically significant

25The random term ε i is of course a function of the initial error term ?i as well as the intervening sequence of

disturbances µ it t T=0, ,K . OLS estimation would require that the error terms are independent of the Zi.

24

(hydrocarbons only at the 10% level). High levels of GDP per capita are associated with theabsence of malaria; low transport costs; a coastal population; and a large endowment ofhydrocarbons per capita. The strong correlation of malaria with income levels is not simply an“Africa proxy.” When we re-run the equation for the sample of countries outside of Sub-SaharanAfrica, in regression (5), we find very similar results. When we enter both Malaria and Tropicarinto the regression (not shown), the effect of the Malaria variable is far more important, suggestingthat the negative effect of the Tropicar variable is largely subsumed by the geographic distributionof Malaria. Of course, these associations are hardly a proof of causality. Not only might theexplanatory variables be proxies for left out variables (e.g. malaria may be proxying for a range oftropical diseases or other liabilities), but there could also be reverse causation, in which highincomes lead to the control of malaria, or to a reduction of transport costs.

In regression (6), we include a vector of variables related to political and economicinstitutions: Socialism, which is a dummy variable for socialist economic institutions; New State,which measures the proportion of time under colonial rule; Public, which measures the quality ofgovernmental institutions; and Open, which measures the proportion of time between 1965 and1990 that the country is open to international trade. We find, in line with many recent studies, thatopenness and quality of public institutions are highly correlated with the level of income. Thesocialist variable is not significant, probably because of the strong collinearity with Open andbecause of the smaller data set once we include the Public variable (since that variable is notavailable for most of the socialist economies). Newly independent countries also do not havesignficantly lower income levels. If the malaria index is not included, the new state variable ishighly significant, which suggests the possibility that the heavy burden of disease in tropical Africaand Asia made these regions more susceptible to colonization. We find, importantly, that bothpolicy and geography variables are strongly correlated with the level of 1995 per capita GDP. Remember that geography may be even more important than suggested by this equation, since thereare reasons to believe that favorable geography plays a role in inducing growth-promotinginstitutions such as open trade and an efficient public bureaucracy. We examine this linkage,briefly, below.

Geography and growth of per capita income

We now examine the forces of convergence and divergence by estimating a cross-countrygrowth equation that allows for the possibility of catching up effects. Thus, we estimate a modelof average annual growth during 1965-90 conditional on per capita income levels in 1965. (Thedates are determined by data availability. For the purposes of the growth equations, we use thePenn World Tables for our measures of PPP-adjusted GDP per capita). We test whether growth isaffected by the initial income level (negatively in the case of convergence, positively in the case ofdivergence), as well as by geographical variables holding constant the initial income and otherpolicy and institutional variables.

We start with a baseline equation similar to those in Barro and Sala-i-Martin (1995), in

25

which average annual growth between 1965 and 199026 is a function of initial income in 1965; theinitial level of education in 1965 (measured by average years of secondary school in thepopulation); the log of life expectancy at birth in 1965; the openness of the economy tointernational trade; and the quality of public administration (regression 1 of table 3). We findevidence for conditional convergence, and standard results for the other variables: output is anincreasing function of education, life expectancy, openness, and the quality of publicadministration. In regression 2 of table 3 we add Tropicar, Pop100km and LDistance. Tropicarand Pop100km are highly significant and of the expected sign. All other things equal, annualgrowth is 0.9 percentage points lower in tropical countries than in nontropical countries.Landlocked countries (Pop100km = 0) experienced 1.0 percentage points slower growth thancoastal economies. Interestingly, the LDistance variable is not significant. This suggests thatdistance to the core may be subsumed by some other variables. Indeed, countries closer to thecore economies tended to be more open and have better public institutions during the period. If wedrop those two variables, then LDistance has the expected negative sign with statisticalsignificance (not shown).

In regression (3) of Table 4, we drop LDistance and add a measure of malaria at thebeginning of the period.27 Initial malaria incidence has a dramatic correlation with poor economicperformance. Countries that had severe malaria in 1966 grew 1.2 percentage points per yearslower than countries without falciparum malaria, even after controlling for life expectancy. Theestimated effect of being in the tropics becomes smaller than it was without the malaria variable,and insignificant. Clearly, the malaria variable is picking up the explanatory power of the tropicsvariable. The effect of initial life expectancy is also reduced, though still large and significant.

The effect of malaria is accentuated if we also account for changes in malaria over theperiod. Countries that had severe malaria, but reduced its prevalence, grew more rapidly thancountries that were not able to control malaria. We use the malaria index for 1994, and calculatethe change in malaria in the intervening period, dMal6694. In regression (4), countries withsevere malaria and no change in malaria are estimated to have grown 2.0 percentage points moreslowly than countries free of falciparum malaria, other things being equal. Countries that hadsevere malaria in 1966 but got rid it of during 1966-1994 period are estimated to have grown 2.5percent faster than countries with malaria, or slightly (0.5 percent) faster than countries withoutmalaria.

In fact, none of the countries in the sample with 100 percent of their land area subject tofalciparum malaria were able to eradicate it completely over this period. The reductions inmalaria were largest in the countries with the least malaria in 1966. Table 4 shows malaria

26Growth is measured as ( ) (ln ln )1 25 1001990 1965⋅ − ⋅GDP GDP .24The malaria index at the start of the growth period 1965-1990 is constructed in a similar manner to the malaria index

for 1994, used above. We digitized a malaria map for 1966 (shown in Figure 5) from the World Health Organization (WHO,1967) to calculate the fraction of a country’s land area subject to malaria. The malaria index is the product of the percent landarea times the percent of falciparum cases in 1990. The mix of falciparum versus vivax and other species of malaria in a givenecozone is not likely to depend very much on vector control or public health measures, nor is it likely to change over time. Forinstance, Sub-Saharan Africa has always had almost 100 percent of the malignant falciparum, while the temperate regions thatonce had malaria had almost no falciparum.

26

reductions specific to different ecozones (Table 4).28 The temperate ecozones were effectivelyfree of falciparum malaria in 1966. The reductions in malaria occurred in the desert (non-temperate) regions, and the subtropics. There was a small increase in the malaria index forcountries in tropical ecozones.

The change in malaria incidence could be partly due to economic growth if growthprovided countries with the economic resources and institutional wherewithal to carry outeffective control programs. To account for the possible impact of economic growth on malariareduction, we instrument the malaria change with four subtropical ecozone variables and twodesert tropical ecozone variables.29 Regression (5) shows that the estimated impact of malaria ongrowth increases when change in malaria is instrumented. There is no indication that fastergrowth is the cause of malaria reduction. Some of the countries that have had the largest increasesin malaria, like India and Sri Lanka, have had steady, if unspectacular, economic growth over thisperiod. Likewise some countries with dramatic decreases in malaria, like Namibia, had almost noeconomic growth. It is only after controlling for other relevant variables that the effect of malariareduction on growth becomes apparent.

The index of malaria, and malaria change, may be more than just a measure of malaria. Itmay be picking up the incidence of other tropical diseases not well indicated by the average lifeexpectancy and tropical area. Among tropical diseases, malaria is widely recognized to be themost important, but the malaria index may also be a proxy for scourges like onchocerciasis,filariasis, and trypanosomiasis.

Malaria occurs throughout the tropics, but severe malaria is heavily concentrated in sub-Saharan Africa, as shown in Figure 4. Africa currently has 90% of the estimated cases of malariaeach year (WHO 1997a), and it is the only region of the world where falciparum malariapredominates. When sub-Saharan Africa is left out of regression (5) (not shown), the size of themalaria coefficients fall to about half their size in the full sample, and the estimates lose statisticalsignificance (although change in malaria 1966-94 is significant at the 7% level). With the loss offifteen sub-Saharan countries from the sample, it is not possible to obtain accurate estimates of theeffect of the tropics and the effect of malaria separately. When Tropicar is left out of the non-African regression, both initial malaria and change in malaria are significant at the 5% level.

In regression (6) we test for agglomeration effects. The basic idea is to see how economicgrowth depends on the scope of the market. A plausible measure of scope of the market is GDPper km2 within the economy in the initial year, 1965. We separate GDP per km2 on the coast andGDP per km2 in the interior, for reasons discussed earlier: population density on the coast is likely

28 Countries in Table 4 are classified by their predominant ecozone from the following groupings: Temperate (temperate,boreal and polar ecozones), Desert (tropical and subtropical deserts), Subtropical (non-desert subtropical), and Tropical (non-desert tropical).

29 The first stage regression of ecozones on the change of malaria is:dMal6694 = 0.01 (0.63)

- 0.19 (0.99) Subtropical Thorn Woodland + 0.12 (1.38) Subtropical Dry Forest- 0.24 (3.29) Subtropical Moist Forest - 1.79 (1.37) Subtropical Rain Forest- 0.55 (1.33) Tropical Desert + 0.46 (0.35) Tropical Desert Scrub

N = 75, R2 = 0.19.

27

to be associated with an increased division of labor and increasing returns, while populationdensity in the interior is likely to be associated with diminishing returns. Note that ln(GDPdensity) = ln(GDP per km2) = ln(GDP per capita) + ln(Population per km2), and since ln(GDP percapita) is already a regressor, we can enter population density or GDP density interchangeablyinto the regression. For countries in which the entire population is within 100km of the coast, weput the interior population density at zero. For countries in which the entire population is fartherthan 100km from the coast, we put the coastal population density at zero. We also drop Pop100kmas a separate regressor, since Pop100km is highly collinear with the two population densityvariables. We use the 1994 measures of Pop100km and the 1965 population levels for the countryas a whole to calculate the population densities in the coastal and interior regions.

The regression estimate is revealing. We now find that higher coastal population densityis associated with faster growth, while higher interior population density is associated withlower growth. Thus, there appear to be economies of agglomeration at play in the coastal regions,though these economies of agglomeration are not powerful enough to overcome the othertendencies towards conditional convergence in income levels. Large populations appear to be anet disadvantage for inland economies which must rely more on their natural base, and have fewopportunities for absorbing population through manufacturing and international trade based incities.

With separate inland and coastal agglomeration effects in regression (6), the estimatedeffect of malaria becomes less precise, slipping to a 7% significance level. On the other hand, ifinitial malaria and malaria change are included as in regression (7), they are both stronglysignificant, but the diminishing returns of interior population density loses statistical significance. We are pushing the limits of the degrees of freedom in our data, but the results suggest that bothmalaria prevalence and inland population concentrations are detrimental to growth.

Our general conclusions from the growth equations are as follows. First, both policy andgeography variables matter. There is no simple “geographic determinism” nor a world in whichonly good policy matters. The tropics are adverse for growth, while coastal populations are goodfor growth. We did not find strong evidence that distance per se from the core markets is animportant determinant of growth. The tropical effect seems to be strongly related to the prevalenceof malaria. This could be the true direct and indirect effect of that disease, or more likely, a proxyfor a range of tropical maladies geographically associated with malaria. The access to coastseems to matter not just in lowering transport costs, but in allowing for some sort of agglomerationeconomies. A dense coastal population is actually seen to be favorable to economic growth during1965 - 90, while a dense interior population is adverse.

If we summarize the implications on a region by region basis, we conclude the following.

Africa is especially hindered by its tropical location; by its high prevalence of malaria; by its lowproportion of the population near the coast; and by the low population density near the coast. Europe, North America, and East Asia, the core regions, by contrast, are favored on all threecounts. South Asia is burdened by a high proportion of the population in the interior, a very highinterior population density, and a large proportion of the land area in the tropics. The transitioneconomies of Eastern Europe and the former Soviet Union, many of which are landlocked, are

28

burdened by a very low proportion of the population near the coast and very low populationdensity near the coast, but these countries are benefited by lack of exposure to tropical disease. Finally, Latin America is moderately coastal, but with relatively low coastal population densities. Also, Latin America has a moderate exposure to the problems of tropics, including the prevalenceof malaria.

In Table 5 we present a decomposition of the growth rates of these regions usingregression (7) in Table 3. We make a statistical accounting of the deviation of each region’sgrowth from that of East Asia during the period. In the case of Africa, health and geographyfactors are estimated to reduce growth by 3.0 percentage points per year, more than the policy andeducational factors. In South Asia, geography is moderately important (-0.8 percentage points peryear), while in Latin America, the geography and health variables explain almost nothing of theshortfall in growth relative to East Asia. As we suggest in the next sub-section, this accountingmay underplay the real role of geography, since economic policy choices themselves are likely tobe a function of geography.

Geographical effects on economic policy choices

We have noted in the theoretical section that geography may affect economic policychoices by altering the tradeoffs facing government. A coastal economy, for example, may face ahigh elasticity of output response with respect to trade taxes, while an inland economy does not. As a result, a revenue-maximizing inland sovereign may choose to impose harsh trade taxes whilea coastal sovereign would not. In this section, we briefly explore this idea with the data at hand,focusing on the choice of openness versus closure to trade in the period 1965-90 as affected bygeography.

The first step is to check the underlying notion: that the responsiveness of growth toopenness actually depends on geography. So far, we have entered Open and the geographyvariables in a linear manner, not allowing for interactions. To check the possibility ofinteractions, we estimate the basic regression equation for three sets of countries: all, coastal(Pop100km >= 0.5) and hinterland (Pop100km < 0.5), and check the coefficient on the Openvariable. Since we lose degrees of freedom in this exercise, we estimate the barebones growthequation, in which annual average growth between 1965 and 1990 is a function of initial income,Openness, malaria in 1966, the change in malaria between 1966 and 1995, and the log of lifeexpectancy in 1965. The results for the Open coefficient are as follows (robust t-statistics inparentheses): All economies (N=92), β = 2.6 (6.8); coastal economies (N=46), β = 3.3 (6.5);hinterland economies (N=46), β = 1.4 (2.5). We see that the growth responsiveness to trade seemsto be more than twice as high in coastal economies.

The next step is to see whether more coastal economies in fact choose more open tradepolicies. This we do by regressing the extent of openness during 1965 - 1990 on the proportion ofland within 100 km of the coast (Land100km), Tropicar, and the initial income level:

Open6590 = -1.12 + 0.23 Land100km – 0.18 Tropicar + 0.19 lnGDP1965

29

(3.2) (2.1) (2.0) (4.5)

N = 106, R2 = 0.44

There does indeed seem to be something to this line of reasoning, though the results are at bestsuggestive, and should be tested more carefully in later work. The early liberalizers, on thewhole, were the coastal economies. This is certainly evident in East Asia, where countries suchas Korea, Malaysia, Taiwan, Thailand, all opened the economy to trade early in the 1960s, muchbefore the other developing countries.

IV. Population Distribution and Economic Activity

The distribution of population around the world is anything but uniform. Large expansesare virtually uninhabited by humans, while almost all the land in Europe and coastal South andEast Asia is tilled or occupied by towns or cities. The dramatic differences in population densityat different latitudes are shown in Figure 7. The geographical features that support high populationdensities seem to fall along two main dimensions. First, there are features that favor denseagricultural settlements, such as soil suitability, inland rivers for local transport and irrigation, andclimatic and ecological systems conducive to rice cultivation (which supports an especially highlabor-intensity of production compared to other grains). Second, there are features that supportmodern economic growth, such as access to the coastline and thereby to international trade. Sincepopulation densities have a very long time dependence, the current distribution of worldpopulation was heavily influenced by demographic trends well before the period of moderneconomic growth. We see from Figure 8, for example, that the high population density regions of1800 are virtually the same as those of 1994 (see Figure 2). Broadly speaking, we can say that theagricultural suitability is more imprinted on global population patterns than the suitability formodern economic growth. Also, the legacy of low population densities in the New World persistsdespite several centuries of in-migration from the Old World.

Very importantly, the geographical conditions propitious for dense agrarian populationsare often very different from those conducive to economic growth. In particular, agriculturedepends more on access to fresh water than on access to the ocean. This has led throughout historyto high concentrations of inland populations that are now substantially cut off from participation ininternational trade. Moreover, as we noted earlier, the dynamics of population change mayexacerbate the biases towards high concentration in inland areas. Rising incomes throughsuccessful industrialization has reduced fertility, making population growth self-limiting in thericher regions. By contrast, the rural areas with poorer growth prospects have some of the highestpopulation growth rates in the world.

The conjunction of a geography that supports high population densities, but not economicgrowth, is the location of the most severe and intractable poverty. Hinterland China, north centralIndia, central Asia, and inland Africa are all far from world trade and dependent on labor-intensive agriculture with significant disadvantages for modern economic growth. Severe endemicdisease burdens, especially in Africa, add to the geographical obstacles. The role of geography in

30

shaping the distribution of population can be seen in the simple population growth identity. Current population depends on the population at some point in the past, and the growth rate duringthe intervening period:

(8) P P ei ir Ti

1 0=

where Pi1 is the current population density in location i, Pi0 is past population density, ri is theinstantaneous growth rate, and T is time elapsed between period 0 and period 1. The populationgrowth rate in each location, ri , depends on geography, as well as initial population. If ri is notallowed to depend on the initial population density, then current population is always exactlyproportional to past population. The population growth rate is given by

r g P Xi i i= +ln ln0 γ

where g and γ are parameters, and X i is a vector of geographical characteristics. Taking logs ofequation 8 and adding an error term ε i ,

(9) ln ( ) ln lnP gT P T Xi i i i1 01= + + +γ ε

We can regress the current population density on an initial population density (say in 1800) andgeographical characteristics, all in logs, using equation 9. The first coefficient will tell us thedegree of persistence in population density. If the coefficient is less than one, g is negative: higherdensity regions grow slower. This is a measure of “convergence” of population density acrossspace, similar to convergence in economic growth equations. The second set of coefficients tell usthe impact of geography on population density given the initial density; that is, the impact ofgeography on population growth in the last two hundred years. If we take the initial point at thetime of the first appearance of humans ( ′ ≅T 500 000, - Diamond, 1997, p.37), then ′ =Pi 0 1 and

(10) ln lnP T Xi i i1 = ′ ′ +γ ε .

The coefficient vector estimated from this specification tells us the unconditional impact ofgeography throughout time on current population densities.

The Results

Equations 9 and 10 are estimated using the following geographical features: accessibilityto the coast and rivers, elevation, malaria, soil qualities and water availability, and ecozones.30 The results are reported in Table 6. There are several clear patterns:

• Being close to inland and navigable rivers is an important predictor of population density, andmore important than being close to the coast.

30 The data sources are explained in the appendix.

31

• Good soils and water supply are important factors in population density.• Population densities are highest in the moist temperate ecozone.• Population density is greater at high altitudes in the tropics, but lower at high altitudes in the

temperate zone.• Malaria has a curious positive correlation with population density.• There is tremendous persistence in relative population density over the centuries, but there is

also some convergence towards a more uniform density. (The coefficient on population densityin 1800 is positive and less than one.)

• Eurasia has much higher population densities after taking into account all the geographicalfactors, and lands of new settlement (the Americas, Australia and New Zealand) have muchlower population density.

The tendency for population to cluster near non-navigable inland rivers, and less on thecoast or near navigable rivers, is directly contrary to relative importance of these waterways forhigh economic output. This pattern, along with the clustering of population in areas of good soiland water (especially the particular conditions suitable for rice cultivation) and in the moreagriculturally productive ecozones, suggests that suitability for agriculture has been the principledriving force behind the population distribution. The distribution of population near rivers ratherthan near the coast is striking when the regressions are done by region (not shown). South Asia,the former USSR, Western Europe, and East Asia all have lower population densities near thecoast, conditional on their distance to rivers. In Western Europe and East Asia, the population hadthe good fortune to cluster near rivers navigable to the sea, while South Asian populations havevery poor access to water-borne trade. Latin America is the only region with a strongerconcentration of population on the coast than near rivers.

Figure 9 uses GIS to identify high population densities that are far from the coast andocean-navigable rivers. The greatest such densities, we see, are in Central Africa, South Asia(especially the Gangetic Plain), the interior of China (with heavy concentrations in the river valleysystems and Manchuria), and Central Asia, including Iran, Iraq, Anatolia, and states near theCaspian Sea.

The much higher population densities at higher altitude in the tropics, unlike the temperatezone, could be due to a less hostile disease environment since many tropical disease vectors arealtitude and temperature sensitive. This doesn’t seem to be consistent with the positive correlationof population density and malaria, though. Looking at regions separately, the population densitiesare significantly lower in malarial areas in all regions but Africa, where population is muchdenser in malarial areas (not shown). Since Africans in malarial areas may have built up partialimmunities to malaria, they may not seek to avoid infection by moving to non-malarial areas whichmay have other disadvantages (e.g. lack of water). Given the strong negative correlation ofmalaria with income levels, though, the higher population density in malarial areas is of courseextremely worrisome.

There may be very long-term geographical arguments for Eurasia’s high populationdensities in comparison with Africa and the lands of new settlement’s lower densities. Diamond(1997) argues that Eurasia’s East-West axis which runs along, rather than across, ecozones, has

32

allowed the movement of crop varieties, ideas, and goods, as mentioned above. Eurasia had thebest selection of native plants and animals, and the lands of new settlement had the least conduciveflora and fauna for original domestication, Diamond has speculated. In any event the new landswere physically isolated from the rest of the world until the modern era, and therefore not part ofthe diffusion of technology, ideas, and trade that gradually permeated Eurasia. Africa, too,suffered from isolation and a North-South axis that hindered the diffusion of Eurasian innovationsand crop varieties.

Population in the lands of new settlement heavily reflects the equilibrium between theeconomic productivity of these regions and the income levels in the European sending countries.Much of the indigenous population was exterminated by conquest and disease soon after firstcontact with Europeans, and since then incomes of the settlers have generally been somewhathigher than income levels of the strata of Europeans considering migration to distant lands. Although income levels in North America, Australia, and New Zealand have remained on a parwith Western Europe, the costs of distance means that the population densities are much lower thanEurope (the Eastern U.S. is the closest to an exception). Africa, on the other hand, resistedEuropean conquest until the end of the 19th Century, largely due to devastating mortality frommalaria and tropical diseases for would-be European explorers and conquerors, despite Africa’smeager military and political defenses.31 With the application of quinine as a malaria prophylaxis,Europe subjugated most of Africa, but few Europeans settled with the exception of the temperatesouthern and northern extremes. Thus the low productivity of the land in Africa was reflected inlow population densities and low incomes.

The mismatch between the geographical suitability for high population densities andgeographical suitability for modern economic growth delineates the regions of mass poverty in theworld. The most intractable poverty is found in the concentrations of poor far from the coast inAsia and Africa. The numbers of poor are much greater in Asia due to higher population densitieson more fertile land. The lands of new settlement have mostly avoided mass poverty sinceEuropeans chose to leave them sparsely settled unless they had the promise of high incomes. Poverty is typical among the surviving native populations, and pockets of dense population thathave not had close ties to a wealthier Europe, such as Haiti, but the numbers are dwarfed by thepoor of Asia and Africa.

V. Future Research Directions and Some Policy Implications

One skeptical reviewer of an early draft of this work said, “Fine, but we knew all this inseventh-grade geography.” We have three responses. First, it is not really true. Seventh-gradegeography did not attempt to quantify the advantages or disadvantages of various parts of the worldin a systematic way, holding constant other determinants of economic performance. Second, evenif true, what was gained in seventh-grade geography is lost somewhere on the way to graduateschool. The vast majority of papers on economic development and growth in the past decade,using the new cross-country data sets and rigorous hypothesis testing, have neglected even the mostbasic geographical realities in the cross-country work. In considerable writing on Africa, for 31 Curtin’s (1989) Death by Migration tallies the deadliness of Africa for outsiders in the 18th and 19th centuries.

33

example, many socioeconomic variables have been tested for their effects on growth, without evenreflecting on the implications of the high proportion of landlocked countries, the diseaseenvironment, the harsh climate and its affects on agriculture, or the implications of low populationdensities in coastal areas. Third, the policy implications of these findings, if the findings are true,are staggering. Aid programs should be re-thought, and the critical issue of population migrationshould be put into much sharper focus.

The research agenda needs to be re-shaped in light of the importance of geographicalvariables. We know precious little about the underlying relationships of climate to agriculturalproductivity, disease vectors, and public health. Not only do we not know the costs of malaria interms of economic development, we barely know the quantitative extent of the disease. Cause-of-death data are not available for most developing countries, and still worse are the data onmorbidity. We lack basic data on transport costs that are comparable across countries, and evenmore important, within countries, between the hinterland and the urban areas. By neglectinggeography variables, we may well tend to overstate the role of policy variables in economicgrowth, and to neglect some deeper obstacles.32

The following four research questions relating to geography bear much-heightenedscrutiny:

• How do transport costs differ across countries? How much of these differences are related topolicy (e.g. port management, poor road maintenance), market structure (e.g. pricing byshipping cartels), or physical geography (e.g. inland versus coastal versus oceanic trade)? How are transport costs likely to change as a result of new information technologies, improvedinter-modal transport, and other trends?

• What is the burden of disease on economic development? What are the channels of effects:direct and indirect costs of infant and child mortality, adult morbidity, premature death? Whatare the main channels of morbidity: direct effects, interactions with other diseases, interactionswith nutrition, etc.? To what extent is the differential burden of disease a result of policy (e.g.the organization of public health services), resources availability for health expenditures, orintrinsic geographic factors, such as the ecology of disease vectors?

• To what extent are observed differences in agricultural productivity a result of: policy (e.g. thetaxation of agricultural inputs and outputs); the quality of inputs; the scope and scale ofagricultural research; and intrinsic geophysical and biological conditions?

• How are fertility decisions affected by geography? Are the high fertility rates in Sub-SaharanAfrica as result of: (1) low population densities in rural Africa; (2) limitations of non-agricultural activities in the hinterland; (3) policy decisions or limitations (e.g. lack ofadequate family planning); or (4) institutional arrangements, such as communal land tenurewhich may lead to externalities in family size?

32On the other hand, since policy variables are often so poorly measured in cross-country work, there is an inherent

downward bias due to measurement errors.

34

Of course, having a better grasp of these issues will only lead to a next step of analysis: towhat extent do transport costs, disease burdens, agricultural productivity, and population growthand density affect overall economic performance? Consider, for example, the relativelystraightforward issue of transport and communications costs. It might be supposed that fallingtransport costs would necessarily favor the hinterland, which is now burdened by very hightransport costs. Krugman, Venables, and others have shown, to the contrary, that a reduction oftransport costs from high to moderate levels can actually disadvantage a high-cost region at theexpense of a medium-cost region, by giving even greater benefits to the second region. Considerour simple setup in equation 3. Suppose that there are two economies with differential transportcosts. Suppose, for example that τi = exp(di µ) where di is the distance of economy i from the coreeconomy, µ is a transport cost parameter that declines over time. Suppose that there is a “near”and a “far” economy, with dn < df . When µ is very high, both economies have zero growth. Whenµ is zero, both have equal and high growth. It is when µ takes a middle value that growth ratesdiffer, with the near economy growing faster than the far economy. Thus, even when we know howtransport costs differ, and how they are likely to evolve, we must have a accurate spatial model tounderstand the implications.

The policy implications of these geographical considerations must of course be informedby clearer research results. Even now, however, we can identify several areas of public policywhich almost surely should be adjusted. First, we should give heightened scrutiny to the specialproblems of landlocked countries, and hinterland populations within coastal economies. There are28 landlocked countries outside of Europe, with 295 million people in 1995. These are, forreasons we have been stressing, among the poorest countries in the world, with an average incomeof $1,673. In a large number of cases, the infrastructure connecting these countries to worldmarkets is seriously deficient. The coastal economies harass the interior economies, or neglect theroad networks that would link them to the coast, or impose punitive effective taxation on transitand port charges. In some cases, there have been heated political clashes between the interiorcountry and the coastal economies. Chile and Bolivia still lack diplomatic relations 119 yearsafter the War of the Pacific cost Bolivia its coastline. Aid programs to improve transportinfrastructure linking landlocked countries to ports almost necessarily require the cooperation ofmore than one country. For example, crucial infrastructure aid for Rwanda includes the repair andmaintenance of the Kenyan road from Nairobi to Mombasa, which transports Rwandan as well asUgandan tea to the Indian Ocean. Such cross-national needs are hard to coordinate and are oftenneglected by country-based donor efforts. By the same token, but perhaps somewhat easier,policymakers should give heightened scrutiny to transport conditions for hinterlands withinnational economies, such as Uttar Pradesh in India, where more than 140 million people liveseveral hundred kilometers from the coastline.

Second, policy makers should examine the likelihood and desirability of large-scale futuremigrations from geographically disadvantaged regions. Suppose that it is true that significantpopulations face local cost or disease conditions that are simply prohibitive of economic growth. The result is likely to be growing pressures for mass migration, first within countries, then acrossimmediate national borders, and finally internationally. We have not yet studied the linkages ofgeography and migration, though it is painfully evident that the linkage is strong. Landlocked

35

countries such as Bolivia have perhaps 15 - 20 percent of the population living in neighboringcountries, especially Northern Argentina. It is estimated that around one-third of Burkinabés areliving in Ghana, Ivory Coast, and elsewhere. In general throughout Southern Africa, there arelarge relatively uncontrolled population movements across national boundaries. While this is along-standing pattern of the region, the consequences are becoming increasingly complex and oftendeleterious, such as the unintended spread of HIV and other diseases and the inability of theenvironmental and public health institutions to cope.

Third, to the extent that the arguments in this paper are correct, they shed a dramatic lighton current population trends. We have shown that future population increases are likely to belargest precisely in the most geographically distressed economies! Consider the United Nationsmedium population projections for the year 2030. The projected annual average growth ratesbetween 1995 and 2030 are mapped in Figure 10. We see clearly that the highest projectedgrowth rates are for the regions that are least coastal, most tropical, and most distant from the coreeconomies. If we regress the projected population growth on geographical characteristics, wefind:

Annual Population Growth, 1995-2030 (Projected, in percentage points) =

-2.1 + 0.91 Tropicar + 0.41 Ln(Distance) - 0.74 Pop100km (3.34) (6.42) (5.31) (4.63)

N = 147R2 = 0.59

We see, for example, that inland economies (Pop100km = 0) have projected growth rates that are0.74 percentage points per year higher than coastal economies (Pop100km = 1). Tropicaleconomies have projected growth 0.91 percentage points above non-tropical economies.

Population pressures in these difficult locations are likely to intensify the pressures formass migration. We therefore require a more urgent look at population policy. A certain calm hasdescended over this policy area, on the questionable grounds that population growth “does notmatter for per capita economic growth.” We have seen the half-truth of this assessment: it may betrue for coastal economies engaged in the international division of labor; it is most likely untruefor the geographically distressed regions where the population increases will be most dramatic.

Fourth, the policy community should re-examine the balance of aid between policy-basedlending to individual governments, which is the currently popular form of aid, and greatlyenhanced aid for basic science on tropical agriculture and tropical public health. The results inthis paper strongly suggest that the tropics are damned not just, or even mainly, by bad policies, butby difficult inherent conditions. If this is the case, the relentless pressures on policy reform may infact be misguided. Perhaps a more effective approach for controlling malaria would do more toimprove the economic environment — and incidentally, improve policy by improving the

36

incentives for good policies that the sovereign faces. There is no doubt that many of the coreissues in tropical health and agriculture are prime examples of international public goods thatrequire a concerted scientific and financial commitment far beyond the means of any individualgovernment. The coordinated agricultural research aid effort is seriously underfunded; thesituation in tropical public health is even more desperate.

37

Data Appendix

Variables used in the Cross-Country Regression Analysis

I. Dependent Variables

GDP per capita:PPP-adjusted GDP per capita, in 1950 and 1990 from Maddison (1995,Tables D1 and F4).

PPP-adjusted GDP per capita, in 1995 from World Bank (1998). Forcountries missing 1995 World Bank data, the data are from CIA (1996) (orfrom CIA, 1997, as noted): Afghanistan; Albania; Bosnia and Herzogovina(CIA 1997); Bhutan (CIA 1997); Brunei (CIA 1997); Cambodia; Cuba;Djibouti (CIA 1997); Equitorial Guinea (CIA 1997); Eritrea; FrenchGuiana (CIA 1997); Iraq; Korea Dem.People's Rep.; Kuwait; Liberia;Libya Arab Jamahiriy; Myanmar (CIA 1997); Somalia; Sudan; Taiwan(CIA 1997); Tanzania; The former Yug. Rep. of Macedonia; Yugoslavia(Serbia/Montenegro). Data for additional countries shown in Figure 1 arefrom CIA (1997).

GDP growth:Instantaneous growth rate of PPP-adjusted GDP per capita from 1965 to1990 from the Penn World Tables 5.6 (Summers and Heston, 1994).

II. Transport Cost and Market Proximity Measures

Lt100km:The proportion of a country’s total land area within 100 km. of the oceancoastline, excluding coastline in the arctic and sub-arctic region above thewinter extent of sea ice (NGS, 1995). Calculated from digital coastlines inArcWorld Supplement (ESRI, 1996a).

Lt100cr:The proportion of a country’s total land area within 100 km. of the ocean orocean-navigable river, excluding coastline above the winter extent of seaice and the rivers that flow to this coastline. Rivers were classified asocean-navigable mainly according to descriptions in Rand McNally (1980),Brittanica Online (1998), and Encyclopedia Encarta (1998). Preciseinformation on our classification of river systems is available from theauthors. Ocean-navigable rivers are displayed in Figure 9. Calculated fromdigital coastlines in ArcWorld Supplement database (ESRI, 1996a) andrivers in the ArcAtlas database (ESRI 1996b).

Pop100km:The proportion of the population in 1994 within 100 km. of the coastline (as

38

defined for Lt100km). The data for population distribution in 1994 comefrom the first detailed world GIS population dataset (seen in Figure 2)described in Tobler, et al. (1995).

Pop100cr:The proportion of the population in 1994 within 100 km. of the coastline orocean-navigable river (as defined for Lt100cr). The population data are asfor Pop100km.

CoastDensity:Coastal Population/Coastal km2 = (Population * Pop100km)/(Land Area *Lt100km). Units: persons per square kilometer.

InteriorDensity:Interior Population/Interior = (Population * (1-Pop100km))/(Land Area *(1-Lt100km)). Units: persons per square kilometer.

Landlocked, not in Europe:Indicator for landlocked country, excluding countries in Western andCentral Europe (Austria, the Czech Republic, Hungary, the FormerYugoslav Republic of Macedonia, Slovakia, and Switzerland). IncludesEastern European countries of Belarus and Moldova.

LDistance:The log of the minimum Great-Circle (air) distance in kilometers to one ofthe three capital-goods-supplying regions: the U.S., Western Europe, andJapan, specifically measured as distance from the country’s capital city toNew York, Rotterdam, or Tokyo.

CIF/FOB shipping cost margin:The ratio of CIF import prices to FOB import prices as a measure oftransport costs from IMF data (Radelet and Sachs, 1998).

III. Other Geographical Variables

Tropicar:The proportion of the country’s land area within the geographical tropics.Calculated from ArcWorld Supplement database (ESRI, 1996a).

Malaria Index in 1966:Index of malaria prevalence based on a global map of extent of malaria in1966 (WHO, 1967), and the fraction of falciparum malaria. The fraction ofeach country’s land area subject of malaria was calculated from digitized1967 map shown in Figure 5 (“limited risk” areas excluded). The intensityof malaria is captured by the fraction of malaria cases that are the malignantP. falciparum species of malaria in 1990 (WHO, 1992). For Africancountries without 1990 falciparum data, we used the WHO (1997b) data(in which almost all African countries with malaria are described as“predominantly” falciparum, which we classified as 100%). The index isthe product of the fraction of land area subject to malaria times the fractionof falciparum malaria cases.

Malaria Index in 1994:Constructed in the same way as the malaria index for 1966, based on a

39

global malaria map for 1994 (WHO, 1997a), and fraction of falciparummalaria in 1990.

Hydrocarbons:Hydrocarbon deposits are the log of total BTUs per person of proven crudeoil and natural gas reserves in 1993 from WRI (1996).

Southern Hemisphere:Indicator for countries wholly below the equator, as well as Brazil,Democratic Republic of the Congo (Zaire), Republic of the Congo,Ecuador, Gabon, Indonesia, and Kenya.

Land Area:Area in square kilometers from World Bank (1997), except for Taiwan andMexico from CIA (1997), with submerged land subtracted out.

IV. Other Economic, Social, and Political Variables

Openness:The proportion of years that a country is open to trade during 1965-90, bythe criteria in Sachs and Warner (1995b). A country is considered to beopen if it meets minimum criteria on four aspects of trade policy: averagetariffs must be lower than 40 percent, quotas and licensing must cover lessthan 40 percent of total imports, the black market premium must be less than20 percent, and export taxes should be moderate.

Public Institutions:The quality of public institutions is based on an index created by Knack andKeefer (1995), which is itself an average of five indicators of the quality ofpublic institutions, including (a) the perceived efficiency of the governmentbureaucracy, (b) the extent of government corruption, (c) efficacy of the ruleof law, (d) the presence or abscence of expropriation risk, and (e) theperceived risk of repudiation of contracts by the goverment. Each countryis scored on these five dimensions on the basis of surveys of businessattitudes within the countries. The subindexes on the five measures are thensummed to produce a single, overall index that is scaled between 0 and 10.

New State:The timing of national independence (0 if before 1914; 1 if between 1914and 1945; 2 if between 1946 and 1989; and 3 if after 1989) from CIA(1996).

Socialism:A variable equal to 1 if the country was under socialist rule for aconsiderable period during 1950 – 1995 based on Kornai (1992).

Life expectancy at birth, 1965:Data from United Nations (1996).

Years of secondary schooling, 1965:Data from Barro and Lee (1993).

40

Urbanization:% of population living in urban areas, 1995, from World Bank (1998).

War-torn:Indicator for countries that participated in at least one external war over theperiod, 1960-85, from Barro and Lee (1994), with additional countriesclassified by authors.

Population:Total population in millions from World Bank (1997).

One degree by one degree population database:

The data for population in 1994 come from the first detailed world GIS populationdataset (seen in Figure 2) described in Tobler, et al. 1995. We aggregated the 5′ by5′ cells to 1° by 1° cells creating approximately 14,000 observations. The worldpopulation distribution in 1800 comes from McEvedy and Jones (1978), mostly ona country-wide basis. The geographical data come from a variety of sources. Incidence of malaria for each 1° cell was digitized from a WHO (1997a) map for1994. Distance of each cell from the coast was calculated from the ArcWorldcoastal boundaries (ESRI, 1992). These boundaries were edited to remove thecoasts north of the winter extent of sea ice as in Lt100km above. Ocean navigablerivers leading to the sea were classified as in Lt100cr as above. Inland rivers arerivers classified as navigable by ArcAtlas (ESRI, 1996b), but with no outlet to thesea, as well as rivers navigable to the sea, but not navigable by oceangoing vessels. Elevation data are derived from the ETOPO world elevation database (NOAA,1988). Land used from rice-growing was derived from the ArcAtlas database onagriculture (ESRI 1996b). Soil depth and stream density (a count of the streams ineach 1° cell from satellite data) come from NASA (Sellers, et al., 1997). Soilsuitability for rainfed crops was derived from Digital Soils Map of the World(FAO, 1995). A classification of land areas into thirty-seven ecozones comes fromthe UNEP (Leemans, 1990).

41

References

Britannica Online. 1998. http://www.eb.com.

Barro, Robert J. 1991. “Economic Growth in a Cross-Section of Countries,” Quarterly Journal of Economics 106(2, May):407-443.

Barro, Robert J., and Jong-Wha Lee. 1993. “International Comparisons of Educational Attainment,” Journal of MonetaryEconomics 32:363-394.

Barro, Robert J., and Jong-Wha Lee. 1994. “Data Set for a Panel of 138 Countries,”http://www.nber.org/pub/barro.lee/ascii/readme.txt.

Barro, Robert J., and Xavier Sala-I-Martin. 1995. Economic Growth. New York: McGraw Hill.

Braudel, Fernand. 1972. The Mediterranean and the Mediterranean World in the Age of Philip II. New York: Harper &Row

Braudel, Fernand. 1981-84. Civilization and Capitalism, 15th-18th Century. London : Collins. 3 volumes.

Caldwell, John. 1982. Theory of Fertility Decline. London: Academic Press.

CIA. 1996. The World Factbook. Washington, D.C.: Central Intelligence Agency.

CIA. 1997. The World Factbook. http://www.odci.gov/cia/publications/factbook/index.html.

Crosby, Alfred W. 1986. Ecological Imperialism: The Biological Expansion of Europe, 900-1900. Cambridge: CambridgeUniversity Press.

Curtin, Philip D. 1989. Death by Migration: Europe’s Encounter with the Tropical World in the Nineteenth Century.Cambridge: Cambridge University Press.

Diamond, Jared. 1997. Guns, Germs, and Steel: The Fates of Human Societies. New York: W. W. Norton.

Encyclopaedia Encarta. http://encarta.msn.com.

ESRI. 1992. ArcWorld User’s Guide and Data Reference. Redlands, CA: Environmental Systems Research Institute.

ESRI. 1996a. ArcWorld Supplement Data Reference and User’s Guide. Redlands, CA: Environmental Systems ResearchInstitute.

ESRI. 1996b. ArcAtlas: Our Earth. Redlands, CA: Environmental Systems Research Institute.

FAO. 1995. "The Digital Soils Map of the World Version 3.5: Notes," CDRom. Rome: FAO.

Gallup, John Luke. 1998. “Agricultural Productivity and Geography,” mimeo, HIID(http://www.hiid.harvard.edu/pub/other/geodev.html).

Grossman, G. M., and E. Helpman. 1991. Innovation and Growth in the Global Economy. Cambridge, MA: MIT Press.

Hall, Robert E., and Charles I. Jones. 1996. “The Productivity of Nations,” NBER Working Paper 5812 (November),

42

Cambridge.

Jones, E. L. 1981. The European Miracle: Environments, Economies, and Geopolitics in the History of Europe and Asia.Cambridge: Cambridge University Press.

Kamarck, Andrew. 1976. The Tropics and Economic Development: A Provocative Inquiry into the Poverty of Nations.Baltimore: The Johns Hopkins University Press.

Knack, Stephen, and Philip Keefer. 1995. “Institutions and Economic Performance: Cross-Country Tests Using AlternativeInstitutional Measures,” Economics and Politics 7(3):207-27.

Kornai, János. 1992. The Socialist System: The Political Economy of Communism. Princeton: Princeton University Press.

Landes, David S. 1988. The Wealth and Poverty of Nations: Why Some are so Rich and Some so Poor. New York: W.W.Norton.

Lee, Douglas. 1957. Climate and Economic Development in the Tropics. New York: Harper for the Council on ForeignRelations.

Leemans, Rik. 1990. “Possible Changes in Natural Vegetation Patterns Due to a Global Warming,” Working Paper WP-90-08,IIASA, Laxenburg, Austria. GIS database available at http://www.grid.unep.ch/matgnv3.html.

Maddison, Angus. 1996. Monitoring the World Economy: 1820-1992. Paris: Organization of Economic Cooperation andDevelopment.

McNeill, William Hardy. 1963. The Rise of the West: A History of the Human Community. Chicago: University of ChicagoPress.

McNeill, William Hardy. 1974. The Shape of European History. Oxford: Oxford University Press.

NGS. 1995. The World: Political. Washington, D.C.: National Geographic Society. Map.

NOAA. 1988. “Data Announcement 88-MGG-02, Digital Relief of the Surface of the Earth.” National Geophysical DataCenter, Boulder, Colorado. GIS database available at http://grid2.cr.usgs.gov/dem/.

Maddison, Angus. 1995. Monitoring the World Economy, 1820-1992. Paris: OECD.

McEvedy, Colin, and Richard Jones. 1978. Atlas of World Population History. New York: Penguin Books.

Murray, Christopher J. L., and Alan D. Lopez, eds. 1996. The Global Burden of Disease. Cambridge, MA: HarvardUniversity Press.

Radelet, Steven C., and Jeffrey D. Sachs. 1998. “Shipping Costs, Manufactured Exports, and Economic Growth,” mimeo, HIID(http://www.hiid.harvard.edu/pub/other/geodev.html).

Rand McNally and Co. 1980. Rand McNallyEncyclopaedia of World Rivers, Chicago: Rand McNally and Co.

Romer, Paul M. 1986. “Increasing Returns and Long-Run Growth,” Journal of Political Economy 94(5):1002-1037.

Romer, Paul M. 1990. “Endogenous Technological Change,” Journal of Political Economy 98(5, Part 2):71-102.

Sachs, Jeffrey D., and Andrew M. Warner. 1995a. “Natural Resource Abundance and Economic Growth,” HIID Development

43

Discussion Paper No. 517a, Harvard University.

Sachs, Jeffrey D., and Andrew M. Warner. 1995b. “Economic Reform and the Process of Global Integration,” BrookingsPapers on Economic Activity 1(August): 1-118.

Sellers, P.J., B.W. Meeson, J. Closs, J. Collatz, F. Corprew, D. Dazlich, F.G. Hall, Y. Kerr, R. Koster, S. Los, K. Mitchell, J.McManus, D. Myers, K.-J. Sun, and P. Try. 1997. An Overview of the ISLSCP Initiative I Global Data Sets.http://eosdata.gsfc.nasa.gov/CAMPAIGN_DOCS/ISLSC/OVERVIEW_DOCUMENTS/overview.html. AccessedJuly 18.

Smith, Adam. 1776[1976]. An Inquiry into the Nature and Causes of the Wealth of Nations. Chicago: University of ChicagoPress.

Summers, Robert, and Alan Heston. 1994. The Penn World Tables, Mark 5.6. http://www.nber.org/pwt56.html.

Tobler, Waldo, Uwe Deichmann, Jon Gottsegen, and Kelly Maloy. 1995. “The Global Demography Project.” Technical ReportTR-95-6, National Center for Geographic Information and Analysis, April. Santa Barbara.

United Nations. 1996. World Population Prospects 1950-2050 (The 1996 Revision). New York: United Nations. Computerdiskettes.

WHO. 1967. “Malaria Eradication in 1966,” WHO Chronicle 21(9, Sept.):373-88.

WHO. 1992. “World Malaria Situation in 1990, Part II,” WHO Weekly Epidemiological Record 67(23, June 5): 169-74.

WHO. 1997a. “World Malaria Situation in 1994, Part I,” WHO Weekly Epidemiological Record 36:269-274.

WHO. 1997b. International Travel and Health: Vaccination Requirements and Health Advice - Situation as on 1 January1997. Geneva: World Health Organization. http://mentor.who.ch/yellow/welcome.htm.

World Bank. 1997. World Development Indicators 1997 CD-ROM. Washington, D.C.: International Bank for Reconstructionand Development.

World Bank. 1998. World Development Indicators 1998 CD-ROM. Washington, D.C.: International Bank for Reconstructionand Development.

WRI. 1996. World Resources Database 1996-97. Washington, D.C.: World Resources Institute.

WTO. 1995. International Trade: Trends and Statistics. Geneva: World Trade Organization.

44

Table 1. Geographical Characteristics of Selected Regions

ContinentGDP/PC

TotalPopulation

Total LandArea

(millionkm2)

Land inTropics

(%)

Populationw/100kmof Coast

(%)

Populationw/100km of

Coast orRiver (%)

LandlockedPopulation

(%)

Distanceto CoreMarket(km)

CoastalDensity

(pers/ km2)

InteriorDensity

(pers/ km2)

Sub-SaharanAfrica 1,865 580 24 91 19 21 28 6,237 40 22

WesternEurope 19,230 383 3 0 53 89 4 922 109 125

East Asia 10,655 1,819 14 30 43 60 0 3,396 381 91

South Asia 1,471 1,219 4 40 23 41 2 5,744 387 287

TransitionEconomies 3,902 400 24 0 9 55 21 2,439 32 16

LatinAmerica 5,163 472 20 73 42 45 3 4,651 52 18

45

Table 2. Level of GDP

(1) (2) (3) (4) (5) (6)lgdp50 lgdp90 lgdp95 lgdp95 lgdp95

(non-Africa)lgdp95

Tropical Area (%) -0.69 -0.99 -0.99(4.13) (5.78) (5.10)

Pop 100 km (%) 0.71 1.00 1.09 0.85 1.21 0.36(4.02) (5.43) (5.27) (3.63) (4.17) (2.53)

Ldistance -0.22 -0.39 -0.34 0.03(2.56) (4.39) (3.41) (0.55)

-2.28 -13.50Shipping Cost(CIF/FOB) (2.32) (4.66)

-1.55 -1.26 -1.15Malaria index 1994(0-1) (6.60) (2.69) (7.65)

0.01 0.01 0.01Log hydrocarbonsper person (1.84) (1.75) (1.85)

Socialism -0.05(0.31)

New State (0-3) -0.06(0.98)

0.23Trade Openness(0-1) (7.38)

0.55Public Institutions (0-10) (3.17)

Constant 9.07 11.19 10.98 10.84 22.64 6.71(13.58) (16.26) (14.10) (9.82) (7.42) (11.06)

Observations 129 129 129 83 52 97R2 0.38 0.58 0.50 0.69 0.56 0.88

Absolute value of t-statistics in parentheses

46

Table 3. GDP Growth

(1) (2) (3) (4) (5) (6) (7)gr6590 gr6590 gr6590 gr6590 gr6590

(TSLS)gr6590 gr6590

GDP p.c. 1965 -2.3 -2.4 -2.5 -2.6 -2.7 -2.3 -2.4(7.70) (8.02) (8.06) (7.87) (7.60) (7.41) (7.09)

0.3 0.2 0.2 0.2 0.2 0.1 0.1Years of secondaryschooling (1.75) (1.77) (1.32) (1.34) (1.15) (0.81) (0.89)

Log life expectancy 1965 6.6 5.5 4.3 3.3 2.4 4.1 3.4(7.23) (6.21) (4.45) (3.60) (1.79) (4.53) (3.89)

1.9 1.9 1.7 1.7 1.7 1.8 1.8Trade Openness 1965-90(0-1) (5.49) (4.79) (4.79) (4.70) (4.39) (4.79) (4.66)

Public Institutions (0-10) 0.3 0.3 0.3 0.4 0.5 0.3 0.4(3.08) (2.63) (3.32) (3.92) (3.66) (3.20) (3.47)

Ldistance 0.0(0.24)

Pop100km (%) 1.0 0.9 0.8 0.6(3.07) (3.01) (2.64) (1.91)

Tropical area (%) -0.9 -0.6 -0.5 -0.4 -0.7 -0.5(2.28) (1.35) (1.09) (0.82) (1.89) (1.44)

Malaria index 1966 -1.2 -2.0 -2.6 -0.9 -1.6(2.15) (3.60) (3.87) (1.86) (2.89)

dMal6694 -2.5 -4.5 -1.9(3.93) (2.12) (2.94)

Log coastal density 0.3 0.2(4.91) (4.34)

Log inland density -0.1 -0.1(2.26) (1.60)

Constant -8.9 -4.1 1.3 5.9 9.8 0.7 4.1(2.90) (1.17) (0.34) (1.57) (1.76) (0.19) (1.08)

Observations 75 75 75 75 75 75 75R2 0.71 0.75 0.77 0.80 0.78 0.80 0.82

Absolute value of robust t-statistics in parentheses

47

Table 4. Level and Changes in Malarial Prevalence between 1966 and1994 by Ecozone

Predominant EcozoneMalaria Index1966 (0-100)

Average Change1966-1994

Temperate(N=57) 0.2 -0.2

Desert(N=23) 27.8 -8.8

Subtropical (N=42) 61.7 -5.0

Tropical(N=21) 64.9 0.5

Note: Countries are classified by their predominant ecozone from the followinggroupings: Temperate (temperate, boreal and polar ecozones), Desert(tropical and subtropical deserts), Subtropical (non-desert subtropical),and Tropical (non-desert tropical). The index and average reduction areunweighted averages over countries.

48

Table 5. Growth Rates in Selected Regions Compared with East Asia

Sub-Saharan Africa-East Asia

South Asia-East Asia

Latin America-East Asia

Growth -4.2 -2.8 -3.6

Explained -3.7 -2.0 -3.0

Initial GDP 1.4 1.0 -1.1

Total -3.0 -0.8 -0.2

Coastal density -0.7 0.0 -0.5 Interior density 0.0 -0.3 0.1 Tropics -0.1 0.1 -0.1 Malaria -1.0 -0.1 0.3

Geo

grap

hy a

nd H

ealth

Life expectancy -1.2 -0.5 0.0

Total -2.1 -2.1 -1.8

Openness -1.2 -1.2 -1.0 Public institutions -0.7 -0.9 -0.7Po

licy

and

Educ

atio

n

Secondary education -0.2 -0.1 0.0

Residual -0.5 -0.8 -0.6

49

Table 6. The Impact of Geography on log Population Density 1994a

(1) (2)log Population density in 1800 0.612

(79.63)**Eurasian continent 1.136

(28.15)**Lands of new settlement -0.998

(23.73)**log Distance (km) to:

Coast -0.040 -0.145(2.26)* (8.13)**

Navigable river to the sea -0.113 -0.188(6.28)** (10.2)**

Inland river -0.324 -0.235(26.48)** (18.4)**

log Elevation in temperate zoneUp to 1000 meters 0.018 0.096

(1.35) (6.77)**1000 to 2000 meters 0.859 1.108

(9.63)** (11.67)**Over 2000 meters -1.361 -0.279

(12.83)** (8.47)**log Elevation in tropics

Up to 1000 meters -0.034 0.034(6.47)** (7.28)**

1000 to 2000 meters 1.801 2.133(6.04)** (6.2)**

Over 2000 meters 0.846 1.296(3.2)** (3.15)**

Malaria (fraction of area) 0.109 2.104(2.48)* (4.72)**

Soil and WaterRice growing land (fraction of area) 1.064 1.335

(10.44)** (13.41)**Soil suitability (0-100) 0.127 0.134

(21.47)** (22.25)**log Stream density (number per cell) 0.144 0.192

(17.77)** (23.42)**Ecozones (compared to Moist Temperate)b

Polar and Boreal -2.273 -3.202** **

Desert -1.224 -1.840** **

Dry Temperate -0.176 -0.409** **

Very Wet Temperate -1.204 -1.539** **

Dry Tropical -0.380 -0.494** **

Wet Tropical -0.932 -1.257** **

Number of observations 13976 14418R-squared 0.74 0.73

Absolute values of robust t statistics are in parentheses. * Significant at the 5% level. ** Significant at the 1% level.

a The regressions included constants that are not reported.b The regression includes 36 ecozones. Moist Temperate ecozone is the left out category. The reportedestimates are the average of ecozone coefficients in each ecozone group. “**” means that all the constituentcoefficients were statistically different from zero at the 1% level.

50

GDP per capita 1995

GDP per capita 1995$450 - 999$1,000 - 2,699$2,700 - 3,199$3,200 - 4,399$4,400 - 4,499$4,500 - 6,499$6,500 - 18,099$18,100 - 21,999$22,000 - 31,100No Data

Tropic of Cancer

Tropic of Capricorn

Figure 1.

51

Persons per square kilometer01-23-56-1011-2021-5050-100100-34530

Population DensityFigure 2.

52

GDP Density

GDP per square kilometer$ 0 - 499$ 500 - 1,099$ 1,100 - 2,999$ 3,000 - 8,099$ 8,100 - 21,199$ 22,000 - 59,999$ 60,000 - 162,999$ 163,000 - 441,999$ 442,000 - 546,000,000No Data

Figure 3.

53

Malaria Index00.00 - 0.070.08 - 0.280.29 - 0.540.55 - 0.810.82 - 1No Data

Malaria Index 1994

Tropic of Cancer

Tropic of Capricorn

Figure 4.

54

Malaria risk - 1946, 1966, 1994Figure 5.

1946

1966

High risk of Malaria1994

Tropic of Cancer

Tropic of Capricorn

55

Figure 6.

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

40.0

45.0E

aste

rn E

urop

e &

Form

er U

SSR

OE

CD

cou

ntri

es

Chi

na

Lat

in A

mer

ica

&C

arib

bean

Mid

dle

Eas

t & N

orth

Afr

ica

Asi

a ex

cl. C

hina

Sub-

Saha

ran

Afr

ica

Wor

ld

Infectious and Parasitic Disease as Percent of Disability Adjusted Life Years (DALYs) from All Causes, 1990

56

Figure 7. Population Density by Latitude, 1994 P

opul

atio

n de

nsity

(per

sons

/ sq

km)

Latitude bands0

20

40

60

80

100

-50 -40 -30 -20 -10 -0 0 10 20 30 40 50 60 70

57

Population Density in 1800Figure 8.

Persons/sq km0.01 - 0.30.3 - 0.7

58

Populations remote from coastline or major navigable river

Tropic of Cancer

Tropic of Capricorn

Figure 9.

Population >500 km0 - 23 - 1920 - 149150 - 399400 - 1,0991,100 - 2,9993,000 - 8,0998,100 - 21,09922,000 - 640,000

Population between 100-500 km0 - 23 - 67 - 4950 - 399400 - 1,0991,100 - 2,9993,000 - 8,0998,100 - 59,99960,000 - 2,100,000

59

Population Growth-0.6 - -0.4-0.4 - 00 - 0.70.7 - 11 - 1.71.7 - 2.32.3 - 3.6No Data

Projected annual population growth 1995 - 2030Figure 10.

Source: United Nations (1996)